[Feature Request] Improve handling of free-tier model rate limiting (e.g. deepseek-v4-flash) to avoid task interruptions

**Background**

First of all, thank you for QwenPaw — it's a great personal AI assistant. We're using it daily with the free deepseek-v4-flash model configured, and overall the experience is excellent.

**Problem**

The free-tier model is frequently rate-limited (429 / limit reached), which causes long-running agent tasks to fail and restart repeatedly. For example, a scheduled or multi-step task may be interrupted mid-way and has to start over, which is quite frustrating and wastes both time and tokens.

We understand this is a free-tier limitation and we're not asking to remove the limits — we'd just like the agent to handle it more gracefully.

**Suggestion**

1. Automatic fallback to a backup model/provider when the primary one hits rate limits (I noticed related discussions in #1327, #4882, #5351 — would love to see this land).
2. For long-running tasks, automatically retry/resume from the checkpoint instead of restarting from scratch when rate-limited.
3. Optionally, surface the current model's rate-limit status in the UI so users know why a task stalled.

Thanks for considering this! Happy to help test or provide more details.

Start by reading the related discussions in #1327, #4882, and #5351, then trace how long-running tasks handle model rate-limit failures. Clarify which of fallback, checkpoint resumption, and UI status is in scope; done should include behavior that prevents a rate limit from restarting the task from scratch.

Scritto dal modello di indicizzazione a partire dal testo della issue.