Table of Persistance.Reference at the end of the stream

[JaroslavTulach]

Pull Request Description

Fixes #9361 by delaying storing of Persistance.Reference instances and creating their table at the end of the stream.

Checklist

Please ensure that the following checklist has been satisfied before submitting the PR:

• All code follows the
  Java,
• Unit tests have been written where possible.

[radeusgd]

I'm really really happy that the reference loops problems is finally getting addressed! ❤️ I will finally be able to proceed with #9361 :)

The logic for resolving the reference loops may not be the most complicated algorithm, but it is non-trivial enough that I think some comments on what was meant are needed - so that it is easier to understand the whole IR caching system that is getting more and more complex.

I'd also suggest considering moving the -1 reference reconstruction to write-time, not read-time, to make this part of algorithm more localized - thus easier to see 'in whole' at once and thus easier to understand.

[hubertp]

I have similar concerns as @radeusgd re lack of documentation for some magic values/array manipulation

[JaroslavTulach]

Thank you @radeusgd for pushing me towards better implementation. Namely, I am very satisfied with d0b0d16 it is clearly better and more easily understandable than the previous version. I believe this PR is ready for integration.