Per-chat sequential queue had no timeout: if a single dispatch hung
(e.g. an agent call that never resolved), every subsequent message in
the same chat stayed `queued` until the gateway was restarted.
Add an optional `taskTimeoutMs` (default 5 min) to `createSequentialQueue`.
After the cap, the in-flight task is evicted from the blocking chain so
newer same-key tasks can proceed. The original task is NOT aborted —
it continues running in the background; we just stop starving the queue.
A warning log surfaces the eviction with the offending key.
`taskTimeoutMs: 0` restores legacy unbounded behavior.
Same-chat FIFO ordering for normal-cadence messages is preserved
(see #64324) — only pathologically slow tasks get evicted.
Fixes#70133.
