Drop overflow cache

[dapplion]

Issue Addressed

Overflow cache is designed to never forget un-finalized not-yet-available blocks.

I believe this is not necessary, and we can drop the complexity to handle that case. Rationale:

Usually a block becomes available on its slot within a second of receiving its first component
over gossip. However, a block may never become available if a malicious proposer does not
publish its data, or there are network issues that prevent us from receiving it. If the block
does not become available after some time we can safely forget about it. Consider this two
cases:

• Global unavailability: If nobody has received the block components it's likely that the
  proposer never made the block available. So we can safely forget about the block as it will
  never become available.
• Local unavailability: Some of our peers will attest to a descendant of the dropped block and
  lookup sync will eventually fetch its components

Even in periods of non-finality, the proposer is expected to publish the block's data
immediately. Because this cache only holds fully valid data, its capacity is bounded to 1 block
per slot and fork: before inserting into this cache we check proposer signature and correct
proposer. Having a capacity > 1 is an optimization to prevent sync lookup from having re-fetch
data during moments of unstable network conditions.

Proposed Changes

Allow DataAvailabilityChecker internal LRUCache to drop the least used entry if it exceeds capacity.

This PR does not change the DataAvailabilityChecker state cache. I think its approach is good as is.

[dapplion]

What's the policy to deprecate DBColumns?

[michaelsproul]

We've never done it before

We could keep it around in DBColumn so it can be used with lighthouse db inspect, and then delete it once the release that removes it is buried under a hard fork

[realbigsean]

added it back here 8366504

[realbigsean]

great docs!

[jimmygchen]

LGTM!

[jimmygchen]

@mergify queue

[mergify]

queue

✅ The pull request has been merged automatically

The pull request has been merged automatically at 880523d