Fixed some datetime types and declared the new SYNTH plugin (but it is commented out as not everything is ready in MeteoIO)

it is also now possible to provide both CSV_UNITS and CSV_UNITS_OFFSET /
MULTIPLIER (see matching commit in MeteIO)

The help texts have been fully rewritten, a new type has been added "COMPONENTS"
in order to clarify what is required to parse date and time by components.
Please note that actually, MeteoIO does not use the CSV_TIMESTAMP key, it is
only there to structure the GUI.

More code replication but more features for the user...

Since default values trigger the printing of the matching keys in the
ini file, explicitely setting default values in SnowpackAdvanced
triggered the writing of a fully populated section in the user's ini
file. This is risky (the user might decide to edit some of these keys
manually) and makes it harder to spot dangerous configuration options in
an ini file. Therefore, the default values have been removed (relying on
Snowpack having them anyway) and are now only provided in the help text
(which also provides an incentive to read the help text).

The Error class was pretty much the first one to exist, and this level
indeed turned out not to be used... it doesn't even really make sense
the way everything works now.
Also cleans up a tiny bit of Qt5 compatibility

The .pro file would set this in the past. Considered switching to
"ifndef NDEUG" but that seems silly, and add_definition but this way
should be trouble-free hopefully and all the code stays the same... feel
free to switch to a more proper way!

If Inishell crashes then the log can be retrieved. This is not possible
of course for users running binaries without compiling, so if we ever
have a use case this could grow into a user setting.

Fixed a mistake: the directory name should be ${CI_COMMIT_TAG#*-} while the filename use ${INISHELL_VERSION}

Key/values read from the GUI would not contain comments (as there is no
way to enter them in the GUI). Now, when an INI file is read the
comments are stored in the key's properties in addition to the value,
and therefore reproduced on output (but still tied to keys).

When multiple panels are loaded for the same key (e. g. via an
Alternative), any arbitrary one would be chosen. Now the correct active
one is hopefully identified.
This should fix issue #893, but not every possible scenario has been
tested...

I think this was only important when the settings file could fail schema
validation. Now that this is done from outside we do not need to be
careful here anymore. Might lose milliseconds but gets rid of a quite
unneccessary function parameter.

Since some special chars are now allowed in sections, the semantics of
the test file did not fit the result...