bug(#3845): introduced roll-bases.xsl

[maxonfjvipon]

Closes: #3845

In this PR I introduced roll-bases.xsl and made it a part of canonical EO and PHI parsing pipelines.
It required:

1. Refactoring TrShaking and moving many XSLs from it to EoSyntax and PhiSyntax
2. Refactoring add-probes.xsl. It led to removing DiscoverMojo from assemble pipeline
3. Refactoring converting XMIR to PHI and to EO
4. Refactoring many yaml test packs
5. Fixing converting to Java
6. Suppressing many lints in eo-runtime

Now we have:

1. We can get canonical XMIR right after EO or PHI parsing
2. Objects bases in such XMIR must be:
   3.1. either ∅
   3.2. or start with .
   3.3. or start with Q
   3.4. or start with $
   all other options are prohibited by XSD
3. all object bases must be rolled into single one until it's not possible anymore
4. canonical XMIR now also assumes that we have named objects only at the first level of abstract objects

[maxonfjvipon]

@yegor256 please check

[yegor256]

@maxonfjvipon this is a massive change, thanks! One question though, about DiscoverMojo. Can't we keep the test intact?

[maxonfjvipon]

@yegor256 I disabled the test because it just does not work with new composite bases. Also we don't need this mojo anymore, it kind of duplicates the functionality of add-probes.xsl and ProbeMojo. That's why it was removed from assemble pipeline. So I can just delete the mojo and the test. WDYT?

[yegor256]

@maxonfjvipon yes, just delete it and its test

[maxonfjvipon]

@yegor256 done, please check one more time

[yegor256]

@maxonfjvipon cool, thanks a lot!