[BUG]Duplicate model_name entries don't fall back correctly — always resolves to second entry, ignores first and third

[YouSummoner]

Quick Summary

When multiple model_list entries share the same model_name (as implied by the fix(config): allow duplicate model_name for load balancing commit), the first entry is skipped entirely and the second is always used — even when the first model works correctly in isolation. If the second entry fails, it errors out rather than falling through to the third.

Environment & Tools

• PicoClaw Version: (0.2.0)
• Go Version: (1.25.7)
• AI Model & Provider: (Google)
• Operating System: (Debian Stable)
• Channels: (Telegram)

📸 Steps to Reproduce

1. Add three model_list entries with the same model_name,
   e.g.:
   { "model_name": "google", "model": "gemini/gemini-3.1-flash-lite-preview", ... },
   { "model_name": "google", "model": "gemini/gemini-3-flash-preview", ... },
   { "model_name": "google", "model": "gemini/gemini-flash-lite-latest", ... }
2. Set agents.defaults.model_name to "google"
3. Send a message — it always hits the second entry
4. Make the second entry invalid — it errors instead of trying the third

❌ Actual Behavior

1. First entry is always skipped
2. Second entry is always used
3. No fallback to third entry on failure

✅ Expected Behavior

Either round-robin across all matching entries, or try them in order with fallback on failure. The first entry should not be silently skipped.

💬 Additional Context

[sunzx]

see bug #975