Make `start_sled_agent` idempotent

Right now, there are two states when the bootstrap agent attempts to start the sled-agent: SledAgentState::Bootstrapping and SledAgentState::ServerStarted. However, to get from Bootstrapping to ServerStarted we call start_sled_agent which is fallible. For many reasons, this method cannot be safely retried. One mechanism to fix this would be to persist the StartSledAgentRequest to the ledger before doing other work and then run code based on what's in the ledger. This would also reduce the latency of adding a sled as we could return to the caller as soon as the request was persisted. It would also make the system crash safe. We handle the bootstore early network config this way, and so it's not so far out there.

Start with sled-agent/src/bootstrap/server.rs at start_sled_agent around line 721, then trace the Bootstrapping and ServerStarted state handling and StartSledAgentRequest. Compare this flow with the bootstore early network configuration's ledger-based handling. Done means retries are safe, the request survives crashes through ledger persistence, and callers can return after persistence.