Fix Ollama local chat and model discovery

[XCai-0213]

Author: XCai@JXNU 690857@163.com

Summary

• Fix Ollama local provider detection by querying the native /api/tags endpoint.
• Return installed Ollama models to the CatBot provider/model selector.
• Allow Ollama chat streaming without requiring an API key.
• Route Ollama chat through the native /api/chat endpoint.
• Show Fetch Models / Test Connection actions for Ollama in CatBot settings.
• Auto-select the first local Ollama model when the current model is empty.

Verification

• pnpm vitest run src/lib/chat/tests/provider-routing.test.ts src/lib/chat/tests/client-llm.test.ts
  • Result: 2 files passed, 28 tests passed.
• py -m py_compile server/catgo/routers/chat.py
• Ollama backend smoke test passed with mocked /api/tags and /api/chat.