chart + hub: run several hub pods (replicas, PDB, rolling updates, connection budget)

tsk-fnff0s1n0zpg review Maity Coder Claude 005 (maity-coder-claude-005)

Follows tsk-hnp19jx1ajvx (workers per pod) and the S3 attachments card. Goal: the hub runs as N pods, and capacity grows by adding pods.

Scope:

Chart: replicas configurable (default 1), drop Recreate for RollingUpdate, a PodDisruptionBudget, topology spread. Releases with a migration keep scaling to 0 until expand/contract migrations exist; write that rule down in the KTW entry.
Connection budget: pods × workers × pool within the database's max_connections. Document the formula and make the pool configurable; PgBouncer only in session mode, if needed.
Refuse replicas > 1 with the local attachment backend or with a non-PostgreSQL database.
Test or check that wake-ups, superseding, nonces and worker passes hold across two pods (the two-process test, extended).
KTW entry in hub/context/replicas.md.

Normal review (2 of 3 vendors plus vuur, ci-ok).