fix(server): reset session status to idle when async prompt fails

[CannonRS]

Issue for this PR

Closes #45610

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

The promptAsync HTTP handler in packages/opencode/src/server/routes/instance/httpapi/handlers/session.ts forks the prompt into a background fiber and only publishes a Session.Event.Error when the prompt fails. The session's SessionStatus entry was never reset in that error path, so the session stays in its last non-idle status (busy or retry) and downstream callers see it as still running.

This PR adds a single statusSvc.set(sessionID, { type: "idle" }) call inside the Effect.catchCause block of promptAsync, wrapped in Effect.ignore so a transient status write failure cannot mask the original error event. The fix mirrors what prompt.cancel already does on the synchronous loop path and keeps the rest of the handler unchanged.

The idle status also publishes the existing SessionStatus.Event.Idle event through the same statusSvc.set path, so subscribers (UI, TUI, plugin hooks) get a consistent transition they already handle.

How did you verify your code works?

• The change is one line in one file: packages/opencode/src/server/routes/instance/httpapi/handlers/session.ts (the only Hunk in git diff upstream/dev..HEAD).
• Manually traced the call path: promptAsync → requireSession → promptSvc.prompt → Effect.catchCause → Effect.forkIn. The new call sits inside the catch block, so it only runs when the prompt fiber actually fails.
• SessionStatus.set({ type: "idle" }) is the same API used by SessionPrompt.cancel and other status reset paths; verified it deletes the map entry and publishes the Idle event.
• bun typecheck runs cleanly in the package after bun install in the worktree (tsgo --noEmit exit 0, against upstream/dev head dc4449df0d + this one-line change).

Screenshots / recordings

N/A — server-side fix, no UI change.

Checklist

• I have tested my changes locally
  • bun typecheck in packages/opencode passes (tsgo --noEmit exit 0) after a fresh bun install in the worktree, against upstream/dev head dc4449df0d plus this one-line change.
• I have not included unrelated changes in this PR

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍