Prevent gaps between the global validator cache and its on-disk representation

[eserilev]

Issue Addressed

Closes #7216

Proposed Changes

Add a new field to the cache staged_indices. When writing to the pubkey cache, we will always insert new validators into the staged_indices mapping. When we write validators to disk, we will generate the database operations from staged_indices. After successful writes to disk, we will flush the validators we wrote to disk from staged_indices.

This should prevent us from ever having gaps between our in-memory validator cache and its on-disk representation.

To be clear, an example of a gap would be:

in-memory cache: [1,2,3,4,5]
on-disk: [1,2,4,5] (3 is missing, but 4 and 5 were written to disk)

If a block were to fail import for any reason, we would still have the new validators included in staged_indices ready to be written to disk on a subsequent block import. If the program crashes before we're able to import the block to disk, we still won't have gaps. On start-up the block will need to be imported to disk anyways which will capture whatever new validators were missed and write them to disk.

With this change #7217 should technically not be needed anymore. But its a simple change and there is no harm in being defensive, so I've chosen not to revert it.

[eserilev]

I think we may be able to write some tests for this change, going to explore that

[michaelsproul]

Going to drop this from v7.0 seeing as the workaround fix seems to work well. We can include this in v7.1.0