[Prover] Prover node keeps track of how many blocks of an epoch were started prior to a crash

[alexghr]

In #10265 we're changing the proving model such that the prover node can keep track of individual blocks in the current epoch. This will allow us to store the proving state and in the event of a crash resume partway through the epoch.

Previously this was impossible because the prover node was forking at the beginning of the epoch and had to sequentially process each block in order to arrive at the correct ending state. With #10265 we can start proving each block independently so we'd only need to figure which blocks of an epoch were not started.

This would also require the orchestrator to restore state: it can no longer use a cache to store intermediate proofs since it wouldn't re-request past proofs.