Scheduled prompts remain queued and do not fire

### Describe the bug

Scheduled prompts in GitHub Copilot CLI did not fire overnight. They appeared to remain queued/active, but no scheduled prompt was delivered back into the session and no assistant turn was started.

This looks like a regression from this week. I expected the scheduled monitor prompt to wake the agent every 20 minutes, but from ~23:04 to ~08:10 local time no scheduled prompt fired.

### Affected version

1.0.71-1

### Steps to reproduce the behavior

1. Start a Copilot CLI session on Windows.
2. Create a recurring scheduled prompt, for example every 20 minutes, intended to monitor a long-running task.
3. Confirm the schedule is listed as active.
4. Leave the session running overnight.
5. Observe that no scheduled prompt is delivered and the agent is not woken by the schedule. In my session, the prompt remained queued/active but no scheduled turns fired.

I saw this with two schedules during the same session:
- an every-15m monitor schedule created first, which appeared queued and did not fire
- an every-20m overnight monitor schedule created later, which also did not fire overnight

### Expected behavior

The scheduled prompt should be delivered at the configured interval and wake/start an assistant turn each time, or it should fail visibly if the schedule cannot execute.

### Additional context

- OS: Windows_NT
- Shell/tooling context: PowerShell-based Copilot CLI session
- Workaround used: a separate PowerShell monitor loop writing to a log, because the scheduled prompt did not wake the agent
- Impact: scheduled monitoring is unreliable for unattended long-running tasks; user expected results by morning and no schedule-fired checks occurred

Commencez par reproduire le prompt planifié récurrent dans une session Windows PowerShell Copilot CLI, en utilisant les scénarios de fonctionnement pendant la nuit ou à intervalles courts décrits dans l’issue. Suivez le moniteur de prompts planifiés et le chemin de réveil de la session, puis vérifiez que chaque intervalle configuré transmet un prompt et démarre un tour de l’assistant, ou signale un échec d’exécution visible.

Rédigé par le modèle d'indexation à partir du texte de l'issue.