implement shard state persistence and recovery upon restart

[raulk]

The DAG store management layer must keep its internal state synchronised with an embedded kv database, injected by the user through the existing ds.Datastore constructor option.

Upon restart, the DAG store must resume its state from the database. On every internal operation, the DAG store must update the shard record in the DB through an upsert.

Data model

• key => shard key
• value => JSON doc (for debuggability right now; later to be moved to CBOR doc for efficiency)
  • mount URL
  • path to transient
  • last state
  • error msg

Resumption procedure

Iterate through kv store, and add a shard for every key, populating it with the persisted data.
For active shards, verify that the index exists.
For failed shards, track them as failed and recover the last known error.

Shards are never resumed into Serving state, and their refcount is always 0.