Download checkpoint data columns from p2p peers

[jimmygchen]

Description

Our current implementation in PeerDAS (#6760) requires the checkpoint server to serve checkpoint blobs via the beacon/blob_sidecars HTTP endpoint.

Once PeerDAS is activated, a full node will no longer be able to serve this endpoint as it won't store full blobs but instead store data columns (samples), and it won't store enough to be able to serve blobs (require >= 50% of all columns to reconstruct blob sidecars) - this means the checkpoint server will need to run a supernode or custody at least 50% of all data columns.

This approach is currently the simplest solution for testing purpose, as the alternative of retrieving columns from p2p peers requires breaking an existing invariant of having all blobs/columns for blocks. However, longer term it would be beneficial to implement the latter, and potentially before PeerDAS goes live.

See this comment for more details: #6760 (comment)

[jimmygchen]

I had a brief conversation with @michaelsproul about this, we think the easiest approach would be just checkpoint sync without the columns, then reimport the block with new columns (potentially using the changes from validator custody).

Validator custody is still WIP, so we could just wait for that before we implement backfilling the checkpoint block. If it's easy to implement once validator custody is done then we may as well implement it, but even without it, missing columns for a single block shouldn't break things too badly.