Time-travel queries against a block hash require that the block is cached #1467
Description
When a GraphQL query specifies at which block hash the query should be executed with block: { hash: "0xdeadbeef"} the current implementation will look in ethereum_blocks to see if a block with that hash exists. If the block is not found, we assume it does not exist, even though it is entirely possible that it exists on chain, and we simply never needed to download it before.
The implementation should make sure that queries can be executed even if they are for blocks that are on chain, but not in our cache.
In practice, this is not a huge issue, since it can generally be assumed that graph-node has cached all blocks from the time it started running as a sideffect of following the chain head.