Improve comptests fork choice generator

[ericsson49]

The PR adds various improvements to the comptests Fork Choice generator:

• randomization improvements for block_tree and block_cover tests
• implemented more consistency checks for block_tree and block_cover tests
• gloas support for block_cover tests
• more robust seed generation scheme
• refactoring of the block_tree main generation loop

Related to #5107

[moodmosaic]

LGTM -- the refactor makes _generate_block_tree much easier to reason about. 👍 Scanned given the size, left a few nits, nothing blocking.

[moodmosaic]

Nice cleanup of the god‑function in _generate_block_tree. 👍

[moodmosaic]

This looks like an improvement to (the old)

attesters = split_list([i for i in range(validator_count)], len(attesting_tips))

That is, before: arbitrary partition of all validators across tips -- after: real beacon committees, then each member picks a random tip.

[ericsson49]

Yes. Additionally, a parent of a tip can be chosen by coin flip.

[moodmosaic]

Do we want to catch all exceptions here and continue? Perhaps catch stale slot, already‑known, etc., and re‑raise everything else(?)

-- or log what was passed when debug=True?

[ericsson49]

I dropped try/catch.
Generally, no exception should be raised here, or they'll manifest themselves in other part of the test.
But it won't hurt to let them be raised earlier.

[moodmosaic]

Perhaps worth swapping the {...} for (...) or [...]?

• {threshold, threshold + 1} -- set, iteration order not guaranteed by the language.
• (threshold, threshold + 1) -- tuple, order is exactly as written.
• [threshold, threshold + 1] -- list, same.

so (in principle) this could affect which value rnd.choice returns for a given seed.

[ericsson49]

Nice catch! Fixed

[ericsson49]

Thanks for the review, @moodmosaic !