Feature request: per-task model selection for scheduled tasks

Problem

Scheduled tasks (created via the scheduled-tasks tooling / "routines") always run on the app's current default model. There is no way to specify a model per task.

Many scheduled tasks are mechanical and repetitive (e.g. run a deploy script, upload a file over scp, verify an HTTP endpoint). Running these daily on a high-cost model wastes tokens, while other scheduled tasks in the same setup (e.g. content generation that benefits from a stronger model) genuinely need a more capable model.

Real-world example from my setup:

• Task A (16:00 daily): generate quiz questions for a kids' learning app -> benefits from a strong model
• Task B (17:00 daily): scp a JSON file to a NAS, run docker compose up -d --build, curl an endpoint to verify -> any small model can do this

Today both tasks silently run on whatever the app default happens to be, and it is not even easy to tell which model a past run used.

Proposed solution

• Add an optional model parameter to create_scheduled_task / update_scheduled_task (same enum as the Agent tool: sonnet / opus / haiku / ...)
• Fall back to the app default when unspecified (current behavior)
• Show the model used for each run in the run history / task list

Why it matters

Scheduled tasks are exactly the workload where cost optimization matters most: they run unattended, daily, forever. Letting users pin cheap models to mechanical tasks would meaningfully reduce token consumption without hurting quality.