Silence embedded etcd "failed to lock file" purge chatter

MIR-1024 noted that failed to lock file fires ~240× in 2h at INFO. It's coming from go.etcd.io/etcd/client/pkg/v3/fileutil/purge.go — embedded etcd's WAL/snap purge routine racing against its own file handles. Harmless but noisy.

Options

1. Wire a configured zap logger into the embedded etcd config that filters this specific message (or raises fileutil's level).
2. Filter at our slog handler layer.
3. Actually investigate why purge is colliding twice a minute — may point at a deeper issue with how we're sizing snapshots or running compaction.

(3) is the "do it right" path; (1) or (2) is the pragmatic quieting move.

Acceptance

• journalctl -u miren in steady state no longer shows failed to lock file.
• If we chose (3), a note in the PR on what was actually happening.