Ensure that indexer blocks are not missed upon restart/crash

[robin-near]

Currently, even if the indexer is started with SyncMode::FromInterruption, it's possible that block updates are sent by the indexer but missed by the MPC node. Upon restart, the indexer is going to yield from the interruption it is aware of, but the missed blocks means that some signature requests would never be indexed. These signature requests would be ignored: if the node is a leader, then it never completes the signature. If the node is a follower but is chosen as a participant, the request would timeout waiting for the indexed signature request.

I propose the following behavior:

• Store a "last indexed final block height" into the DB; commit it at the same time as committing signature requests in a block (also requires committing signature requests of a block in a single commit).
• Upon startup, read the last indexed final block height, and ask the indexer to start yielding blocks from there.
• Signature computation should include the SignatureId as well as the block height from which the signature request was indexed.
• When participating in a signature computation, if the signature is not found while the "last indexed final block height" already passed the specified block height, the computation is aborted rather than waiting forever.