Clarify priority of Tycho-mappings

[HannesWell]

As discussed in #1264, this introduces a well defined priority for all of Tycho's polyglot mappings.
The exact order is:

1. pom.xml
2. bundle (i.e. META-INF/MANIFEST.MF)
3. feature (i.e. feature.xml)
4. repository (i.e. category.xml and *.product files)
5. target (i.e. *.target)
6. aggregator (i.e. a directory with one of the names specified via system property tycho.pomless.aggregator.names)

Furthermore this updates to polyglot-common 4.9 to use takari/polyglot-maven#242 and introduces a PomXMLMapping. The latter is necessary because the default org.sonatype.maven.polyglot.mapping.XmlMapping erroneously claims to handle feature.xml models (because of their ending). Without a dedicated PomXMLMapping we would have to move the XMLMapping as second last in the list above and would have to check in each other Tycho-mapping if the project also has a pom.xml.

[github-actions]

Test Results

324 files  324 suites   2h 14m 6s ⏱️
290 tests 286 ✔️ 4 💤 0 ❌
580 runs  571 ✔️ 9 💤 0 ❌

Results for commit 4ac3d1b.

♻️ This comment has been updated with latest results.

[HannesWell]

I split the this change into two commits.
The first one only introduces the mapping-specific priorities (the actual purpose of this PR), the second one applies some clean-up/code re-ordering.
If you prefer I can open the second commit as separate PR, but the commits can also be reviewed one by one.

[laeubi]

Just a minor comment before we can merge this, maybe you also want to squash all and use one commit message that covers all changes.

[HannesWell]

Just a minor comment before we can merge this, maybe you also want to squash all and use one commit message that covers all changes.

I would prefer to keep the changes in distinct commits. Especially the last change to NIO Path changes many places and the other changes which would make it harder to find the changes of priorities if one later goes through git.
But the specification of distinct priorities plus method reordering and Set simplifications can be squashed without making the result too complicated.