Make it possible to read only a subset of available collections in readers

[tmadlener]

BEGINRELEASENOTES

• Add a collsToRead argument to readEntry and readNextEntry for the ROOTFrameReader to allow limiting the collections that are actually read.

ENDRELEASENOTES

This is an (early) proposal for introducing this functionality which would in the end address #499

• Interface design: Do we want to have the possibility to change this frame-by-frame?
  • Alternatively would have to introduce a configuration method that takes a category name and a list of collections.
• Should we only pass on the limited collection id table?
  • Interplay with Frame and assumptions there
• Add tests and checks to avoid breaking things when collection names are passed that are not even available from the data
  • The way the check is implemented only names that exist in the frame will be checked. Non-existent names in collsToRead will be silently ignored.
• Same capabilities for RNTuple reader
• Documentation (once the rest has settled)
• Ensure that files that have been read with a limited set of collections and are then written again, can be read without issues.
• Needs Require C++20 and update to C++20 #698 as it also starts using std::ranges.
• Needs Make sure RNTupleReader builds with ROOT > 6.34 #719 to pass on dev3 stacks
• Needs Refactor the read tests partially #724
• Includes [ci] Switch to LCG_106b for sanitizer workflows #730

[tmadlener]

Revived this as this is likely to become an issue for moving EDM4hep forwards otherwise. I have settled on making it possible to pass the collections that should be made available on a frame-by-frame basis as that is the most flexible and is also rather simple to implement.

I still have to make sure that the idTable that is available from the constructed Frames is properly cleaned up and also add some tests that passing in unavailable collection names doesn't break the whole thing.

[tmadlener]

It looks like we don't strictly need to trim down the collection id table, as it is by now really only used for the lookups between hashes and collection names. I will keep it as is and leave trimming that down for now as an optimization that can also be done later.