Decouple the collection buffer creation from the collection

[tmadlener]

BEGINRELEASENOTES

• Introduce a CollectionBufferFactory that can create the necessary buffers from a collection type, a schema version and a subset collection flag.
  • Use this factory throughout all existing Readers
  • Remove createBuffers and createSchemaEvolvableBuffers from podio::CollectionBase interface
• Make the minimum allowed schema_version 1 in the yaml definition files. Default to 1 if no schema_version is provided
• Add a schemaVersion to the DatamodelDefinition.h header that is generated and that can be accessed via {{ package_name }}::meta::schemaVersion. Use this to propagate schema information to the necessary places.
• Make SIOBlocks write the current schema version, such that on reading they can generate the appropriate buffers for the version on file.

ENDRELEASENOTES

This PR continues the cleanup of internals and the introduction of full schema evolution capabilities. The most important bit is that it completely disentangles the buffer creation from the collection interface, so that it is now possible to create buffers without having to create a collection first. This is achieved via a CollectionBufferFactory (singleton) that is populated during datamodel library loading. The rest of this PR is mainly shuffling some parts of existing code around and removing some other bits that were previously necessary to "steal" the buffers from the collection they were created from.

This does not yet address any schema evolution concerns other than providing the possibility to generate different buffers for different schema versions and the basics for populating the factory with versioned creation function. For actual schema evolution functionality this has to be populated accordingly (either by users or via automatic code generation).

@hegner this introduces an other de-facto registry, but I don't see an easy way around that. We could hide the fact that it is a singleton registry behind an interface of global functions if we wanted to.

This should solve the issue that is currently blocking #257.

TODO:

• Proper writing of schema version for SIO backend (should be possible to use the version mechanism of SIO here)
• Documentation

[tmadlener]

Updated the docstring for the CollectionBufferFactory to include the basics and some of the assumptions. Since this is a class that is mostly targeted at internal use, I have not created additional documentation in the user facing markdown files.

[hegner]

@tmadlener - did you see the cling error of the IBs before?

[hegner]

Thanks. To me it looks fine.
There is only a nitpicking one about underscore vs camel case for variable names. But that we should discuss for coding conventions anyhow.