F-130: C1 recovery-mode admin HTTP surface

[fdatoo]

Summary

• Adds ReplayError typed error to internal/eventstore so replay failures carry the failing position and projector name, extractable via errors.As
• Adds RecoveryProvider interface in internal/observability/ with five HTTP handlers: GET /events, GET /projection-cursors, GET /skipped-events, POST /events/{position}/skip, POST /shutdown
• Implements RecoveryProvider on *Daemon; all five endpoints guarded by InRecovery() — return 404 when daemon is not in recovery phase
• Adds 30-second heartbeat goroutine in enterRecovery that logs failure reason + position until daemon context is cancelled
• Wires shutdownCancel on Daemon so the POST /shutdown handler can cancel the daemon's context and trigger a clean exit
• Adds ProjectorNames() to eventstore.Store so the skip handler can validate the projector field against registered projectors

Test plan

• go test ./internal/eventstore/... — TestReplay_ReturnsReplayError, TestReplay_SkipEventAllowsReplayToProceed, TestStore_ProjectorNames
• go test ./internal/observability/... — 20 handler tests covering not-in-recovery (404), success paths, empty slices, bad input, unknown projector, shutdown
• go build ./... — clean compile

🤖 Generated with Claude Code