add presence feed to cluster mode via broadcast to all nodes

[fabracht]

Summary

Cluster-mode presence feed, behind the same opt-in --presence / MQDB_PRESENCE flag as agent mode.

• each node publishes retained presence for its own clients to $DB/_presence/{client_id} and broadcasts it to every other node, which publishes it locally
• cross-node delivery uses a broadcast rather than publish routing because TopicTrie::match_topic returns nothing for $-prefixed topics, so a $DB/_presence/# subscriber can never resolve as a remote target
• _presence is added to the $DB/ pass-through list so the message reaches subscribers instead of being taken for a database operation
• live connections are tracked per client id so a session takeover does not report a still-connected client as disconnected
• adds mqdb dev test --presence covering same-node, cross-node both directions, retained state for a late janitor on another node, and a rejected forgery

Every node ends up holding the retained set, so a janitor sees current state from whichever node it attaches to.

Test plan

• cargo make format-check clean
• cargo make clippy passes with zero warnings
• cargo make test: 24 suites, 1196 passed, 0 failed
• cluster unit tests: local publish plus broadcast, internal-client and takeover suppression, received broadcast published locally
• live 3-node cluster mqdb dev test --presence: 7 passed, 0 failed