Fix BestFinalized method

[farazdagi]

What type of PR is this?

Bug fix

What does this PR do? Why is it needed?

• When writing tests for init-sync, interesting bug was discovered (introduced in Fix 2nd Stage of Sync #7006):
  • When peers are selected, peers with higher epochs are preferred (i.e. more knowledgeable peers are preferred).
  • During Medalla incident we wanted to make sure that head slot reported by peers also affects sorting.
  • Sorting wasn't implemented correctly: peers having the same head slot but different epoch were compared using

pidEpoch[potentialPIDs[i]] > pidEpoch[potentialPIDs[j]] && pidHead[potentialPIDs[i]] > pidHead[potentialPIDs[j]]

Which is always false in that case i.e. peer epoch is being ignored, producing almost random data (depending on order items are fed into slice.Sort, which is arbitrary).

• What makes it a bit worse: if peer with lower epoch has higher head (happens quite often in period of non-finality, especially), another peer with higher epoch will not be ranked higher (as that conditional above will be false).

Which issues(s) does this PR fix?

N/A

Other notes for review

• Since we have a separate BestNonFinalized() function (where sorting is done by head slot), which is used in init-sync for syncing slots coming after the finalized checkpoint, BestFinalized() should always use epoch as the main factor when selecting peers (head slots only adding to that).
• I've rewritten test, to rely on test vector:
  • Preserved the original test case
  • Added regression test
  • Added couple of more tests that might be interesting (including checking that max limit of returned peers is enforced indeed - that was untested)