Stuck threads self-heal — version forks now reconcile instead of wedging

Previously, if a node's durable storage row ended up on a newer version lineage than the live hub serving it (for example after a reactivation seeded from a stale cache), every write the hub made was refused by the storage write guard — silently. Chat threads hit by this froze forever: the status never left "executing", new messages queued but never ran, and even the built-in recovery watchdog's writes were refused.

Now the owner detects a refused write, adopts the durable truth, and continues from there — the very next write lands and the thread (or node) recovers on its own. The version stamping that manufactured these forks after a reactivation has also been fixed at the root, so the situation arises far less often in the first place.