Scheduled jobs fail after long runs with `Proxy request failed`

[Femumme]

Scheduled jobs running for roughly 10–15 minutes can fail with:

{"error":"Proxy request failed"}

On v0.16.1, schedules submit through synchronous POST /session/:id/message, while completion monitoring starts only after that request returns. A connection interruption or OpenCode restart therefore fails the run.

Could schedules use POST /session/:id/prompt_async and then start existing completion monitoring? This would avoid keeping submission request open for full run.