Ingest historical block hashes

[leoyvens]

An issue affecting the network, which sends queries by block hash, is that graph node will fail if it cannot find the block hash in its block cache. The block ingestor will ingest blocks within the reorg threshold, but holes can happen and nothing is done to fill them.

Since historical queries mean that all block hashes are potentially necessary, a way to solve this is to run a backfill ingestor job that ingests all historical hashes.

In terms of implementation, we'd want to be able to have only the block hash and number rather than the whole header, which is not valid in the current block cache structure. Also we'd probably want to leverage the firehose and not need a json-rpc implementation of the backfill.

[lutter]

This would also address #1467

[azf20]

I think this is supported for Firehose Blockstreams @maoueh ?

[leoyvens]

After #3738 we'll also need to ingest the timestamps to continue supporting the _meta field.

[azf20]

We could alternatively lazily fetch blocks when there is a query (similar to #4768 )

[madumas]

A note only to point out that this is a pain point on the network. Indexers may be forced to purge the block cache, and there is no easy way to "refill" it. The lazy fetch solution would be great.