RGB is a #RRGGBB hex string, not a R,G,B
Current ltSaveDelay missing from the LIGHTS page

resolve #2451

Paranoid locking doesn't really matter in the current setup (but, leave some comments about how to solve it when it actually does).
Also copy internal tuple of save+transition+report and give the copy to the callback. Thus, don't assume the consumer is always nice,
and does not declare refs to internal struct variables, making this kind of hard to track.

Refactor internal code so it does not export functions to the built .o

Rephrase a bit, since it is a yes / no question box.
Losing the connection triggers ping handlers anyway.

per https://docs.npmjs.com/cli/v7/commands/npm-exec#compatibility-with-older-npx-versions
using --no / --no-install to avoid installing latest versions

making sure ci_install handles that instead, since the specific version in set via the package.json
(hopefully, npm docs actually mean that this will work)

Refactor WebUI:
- remove jquery dependency from the base custom.js and use vanilla JS
- remove jquery + jquery-datatables dependency from the RFM69 module
- replace jquery-datatables handlers with pure-css table + some basic cell filtering
  (may be incomplete, but tbh it is not worth additional 50Kb to the .bin size)
- introduce a common way to notify about the app errors, show small text notification
  at the top of the page instead of relying on user to find out about errors by using the Web Developer Tools
- replace <span name=...> with <span data-settings-key=...>
- replace <div> templates with <template>, disallowing modification
  without an explicit DOM clone
- run `eslint` on html/custom.js and `html-validate` on html/index.html,
  and fix issues detected by both tools

Streamline settings group handling in custom.js & index.html
- drop module-specific button-add-... in favour of button-add-settings-group
- only enforce data-settings-max requirement when the property actually exists
- re-create label for=... and input id=... when settings group is
  modified, so checkboxes refer to the correct element
- introduce additional data-... properties to generalize settings group additions
- introduce Enumerable object to track some common list elements for
  <select>, allow to re-create <option> list when messages come in
  different order

Minor fixes that also came with this:
- fix relay code incorrectly parsing the payload, causing no relay names
  to be displayed in the SWITCHES panel
- fix scheduler code accidentally combining keys b/c of the way C parses
  string literals on separate lines, without any commas in-between
- thermostat should not reference tmpUnit directly in the webui, replace with
  module-specific thermostatUnit that is handled on the device itself
- fix index.html initial setup invalid adminPass ids
- fix index.html layout when removing specific schedules

Sync changelog with recent changes
Codacy is no longer running (...pending viable config for clang-tidy, etc.)
Some of the features were only in wiki

Additional div made the mutator think we still have some active elements.
Also, small clean-up for type consistency in the WS part.

even when the cfg object does not list it, like with scheduler button
trying to fix #2459

since some of them are allowed to yeild(), don't change things while it is already running

Build with Core 3.0.1 release and update the git toolchain to match the upstream

fix #2457

Amends 7ea73554

- generic mdns query function so we don't forget about the removeQuery()
- full handle mdns logic in the mqtt module, check for running mdns
instance and periodically poll for _mqtt._tcp service
- autoconnect only when MQTT is enabled *and* there is no server
- use mqtt::settings and mqtt::build namespaces for settings
- some refactoring for topic <-> setting application
- prefer const String& to const char* in topic generation

ref. #1033, this could be used as-is with a generic analog sensor code
by substituting ::analogRead(channel) with I2CPort::read(channel)

And update other options to support y, yes and true

Do not install libraries globally in CI, use the specified PIO .ini option.
Similar to the other package installers, prefer package cache to the actual library storage contents.

More concise foreach, generic template code also no longer expands down to ::_read_kv
Also helps with needless inlining happening with recent gcc versions

Quoting gcc-4.8.2

> espurna/sensors/BaseAnalogEmonSensor.h:19:7: note: 'BaseAnalogEmonSensor' is not literal because:
> class BaseAnalogEmonSensor : public BaseEmonSensor {
>       ^
> espurna/sensors/BaseAnalogEmonSensor.h:19:7: note:   'BaseAnalogEmonSensor' has a non-trivial destructor

Ratio, Mains are specific to a magnitude type.
Make keys use global index, and do a (temporary) WebUI workaround
to use 0 index instead of the global key.
Migrate to the new settings keys.

Local magnitude-specific index is still in play, pending changes in the rest of sensors
(ADE7953, PZEM004T{,V30}, ...) to remove it completely.

Local sensor index could still be useful though, to help out with some settings
in the future as we don't really enforce magnitude type uniqueness within sensors.

Locks the value to the latest one for a bit. Another option would be
to yield() / delay(...) to spend the time somewhere else.

- fix dubious pinMode(0, INPUT), which does not work with esp8266 as A0 != GPIO0
- I2C sensor locking code can now exist separately from the base class
- EmonSensor -> BaseAnalogEmonSensor
- move AnalogEmon-specific methods from BaseEmonSensor to BaseAnalogEmonSensor
- clean-up protected member usage across the base, force the class to implement certain methods instead
- more code reuse between sensor versions
- drop local magnitude indexes & shared arrays of values in favour of separate
class instances, refactor ADS1X115 to the new model
- (techically) allow ADS1X115 code to be used with more than one i2c address
by having a shared port instance

Remove 2.3.0 code for ldscripts
Update

3.0.0 now uses macro from esp-idf
avoid deprecation warnings for the source here
(still shows it with the libs though, when building without -w)

Make sure it does not have global flags related to the HTTPCLIENT
Use generic ::updateFS with updater, and also support generic "fs" key
in the server response

Should be buildable once again

Restore the behaviour from #1468, where button clicks were triggered
Inject relay provider using the relayAdd, add buttonAdd to provide input handling

Remove extra button handling code from the WebUI

Still assuming that DUAL code really needs a syncronized way to handle
input & output with relays, so we can't treat it as a real button.
Without the device it is only guessing, and various issues / comments
are not really clear on that part :/

Leave build.sh as a local script, assume that CI will use the release.sh instead
Generator parser returned None at the end (...this is not rust...), also
fix args.* typos with a more declarative structure.

- remove version handling, PIO already does this for us
- directly set generate_release_sh.py arguments instead of having a
getopt proxy for each one (and having to deal with short-arg support)
- slightly more logs for the build process

Multiple ways to specify version string through environment variables:
- `ESPURNA_BUILD_FULL_VERSION` to set full version string
By default it is empty, and the version is combined using the values specified below
- `ESPURNA_BUILD_VERSION` to modify the first part of the version string (1.15.0-dev)
By default, uses espurna/config/version.h APP_VERSION value
- `ESPURNA_BUILD_REVISION` to specify revision part of the version string
For example 12345678, which is expanded as either .git12345678 or -git12345678
(depending on whether the version string contains hyphen)
By default, use git to retrieve the first 8 characters of the current HEAD SHA value
- `ESPURNA_BUILD_VERSION_SUFFIX` to specify build metadata part of the string
For example nightly20210607, which is added to the full version as 1.15.0-dev+nightly20210607
Empty by defauld

Adds -t build-and-copy which uses the values above to copy firmware.bin, configurable with:
- `ESPURNA_BUILD_NAME` to set the suffix of the filename.
By default, uses the $PIOENV (aka the string after the env: in the .ini file)
- `ESPURNA_BUILD_DESTINATION` to specify where to copy the .bin files
By default, uses $PROJECT_DIR

Resulting file is stored at:
${ESPURNA_BUILD_DESTINATION}/${ESPURNA_BUILD_FULL_VERSION}/espurna-${ESPURNA_BUILD_FULL_VERSION}-${ESPURNA_BUILD_NAME}.bin

In addition, modify generate_release_sh.py to use the new environment variables.

- make use of scheduler::settings when listing schedules
- do not be overzealous with settings validation & parsing
- cleaner wildcard() usage, we return a String not a ref to something internal.
but, it is techically allowed to use `const T&` for a `T` temporary which will be destroyed
after the apiRegister() callback returns

Restore was unintentionally broken after switching from per-relay to
per-schedule setting. Instead of using schedules count it used the
number of relays (or curtains), and never light channels.

Instead, loop over available schedules and check the 'restore' flag,
calling the action when the schedule was supposed to be active.
Similar to the old approach, check for the most recent action and call
it (also, discarding the same type & target pairs, to avoid scheduling
conflicting actions)

time_t seconds offset is refactored into tm.tm_mday offset.
this causes the function to only work with local time, however
it *should* correctly offset day-by-day in case it somehow stumbles
upon dst transition

For settings:
- schSwitch renamed into schTarget
- schTarget default is no longer 0xff, but 0
- schType is a mandatory setting, instead of 1 (switch) default to 0 (none)
- move getSetting into functions to reduce code size
- delSetting case moved into the initial block that runs before restore() and check()

Plus, refactor C-style naming into namespaces similar to the other modules

Buffer size of 16 is not enough for the output, bump to 32

Specific bug can be triggered by adding >16 relays and using RelayMask
Since the used API does not check buffer size at all, it ended up
overwriting the stack just before the callsite. e.g. relay boot would
show relay count as 48 (aka '0')

Remove explicit reserve() as well, buffer concat will take care of that