Some persistence callers have no deadline, so they stay blocked on a silent Postgres connection #12449

Follow-up to #12246, outside the shard io semaphore that #12441 covers.

Expected Behavior

Persistence calls end once a bounded deadline passes.

Actual Behavior

Three background callers pass no per-call deadline, so on a connection that goes silent without a reset they can stay blocked on either driver:

• nsregistry.(*registry).refreshNamespaces (starts from context.Background())
• ringpop.(*monitor).upsertMyMembership (lifecycle context)
• matching taskManagerImpl.CreateTasks (WithCancel(Background()))

In my #12246 repro with postgres12_pgx, the matching writer still stalled and workflows stopped. With lib/pq, calls that do have a deadline block too (lib/pq#620).

Steps to Reproduce the Problem

1. Run the server against Postgres through HAProxy with a steady workflow load.
2. Black-hole established client connections without a reset.
3. Dump goroutines after a few minutes.

Specifications

• Version: main 5a78ff3 (lib/pq); earlier Unload a history shard when its io semaphore is stuck #12441 build (pgx, unload off)
• Platform: macOS arm64, Docker Desktop, Postgres 16

I'd add per-call deadlines to those three callers and a docs note recommending postgres12_pgx for PostgreSQL. OK to send a PR?