This PR fixes a couple of bugs in event-replay error handling:

• the serf dropping the bail mote and producing only a stack trace where the king expected both
• there was an off-by-one error in interpreting the error response (the king behaved as if event N failed, where it was actually event N+1)

These bugs have been present since the IPC protocol rewrite in mid-2020. They're minor, in that replay errors are all currently fatal. But they're major, in that they preclude accurate error reporting, which is essential in determining if the error is (manually) recoverable (ie, was the failure deterministic?).

NB: errors on replay should (but do not yet) include context-dependent instructions for recovery procedures, or instruct the pilot to contact [email protected].