server: improve handling of ambiguous liveness writes on decommission

**Describe the problem**

From #109883 [thread](https://github.com/cockroachdb/cockroach/issues/109883#issuecomment-1746029559):

> If a liveness KV write with the DECOMMISSIONED membership status for the "self" NodeID is ambiguous, the write is unknown to have failed, or to have succeeded. If the ambiguous write failed, retrying is totally idempotent and allowed - the retry can succeed, and the status can be returned just fine. If the ambiguous write succeeded, however, a retry is not allowed, as KV API usage is not permitted once a node's DECOMMISSIONED status has propagated through gossip and has been written to the decommissioned node tombstone storage. This is a real bug, though in practice it is relatively harmless (only affecting status reporting to the CLI) and I'd imagine exceedingly rare outside test scenarios, so I'm not sure this warrants the release-blocker label.

Jira issue: CRDB-32680