Properly handle skip slots in fork choice

[djrtwo]

Due to the way epoch pairs are justified/finalized and how we handle skipped slots in the consensus, we have to be careful to ensure not only that the justified/finalized block is in the chain, but that it is in the chain at the precise justified/finalized epoch (accounting for skip slots in exceptional cases)

Previously I erroneously put in conditions such the following:

assert block.slot > compute_start_slot_at_epoch(store.finalized_checkpoint.epoch)

This just ensured that the block was in fact later than the finalized epoch, but not that the finalized root was in the chain (potentially at a skip slot) at that epoch.

To fix this, we modify get_ancestor to return the latest root at or before the queried slot in the chain in question. This mimics the behavior of how we save roots in state and how we vote on roots, and it very nicely fixes this corner case.

---

Helps properly handle cases in which we justify/finalize blocks after skipped epochs like in the following:

Say block 64 at epoch 2 is finalized. Blocks that build off of block 64 at epoch 1 should not be considered in the fork choice. the modificaiton to get_ancestor protect against this case