Optimize backFill range finder query

[i-norden]

Current method for finding missing headers takes too long once the database grows large. Rather than finding entire ranges of blocks that are missing and then fetching chunks in batches, have multiple workers concurrently SELECTing single block heights where a header is missing and fetching statediff payloads at those heights.

[i-norden]

The separate, ongoing backfill process is now deprecated. We now do a lazy backfill approach, only backfilling when there is a cache miss on an incoming request.

This will cause some problems with certain operations, namely state operations where it is unknown what block contains the diffs that contain the missing nodes required for the operation.