feat: Add LLM connection test and model discovery

[buptcws-cmd]

Summary

• Add "Test Connection" button for LLM configuration that validates API credentials and fetches available models
• After successful connection test, model selection shows dropdown instead of text input
• Support for OpenAI, Qwen, DeepSeek, Moonshot, Ollama, and custom OpenAI-compatible endpoints
• Claude/Anthropic excluded from connection test (native API not OpenAI-compatible)

Changes

New Files

• pixelle_video/utils/llm_connection.py - Connection test and model discovery utility

Modified Files

• pixelle_video/llm_presets.py - Add connection test fields (supports_connection_test, models_url, models_response_path, model_id_field, requires_api_key)
• web/components/settings.py - Add Test Connection button and model dropdown UI
• web/i18n/locales/en_US.json - Add i18n translations
• web/i18n/locales/zh_CN.json - Add i18n translations

Features

Provider	Test Connection	Model Discovery	API Key Required
OpenAI	✅	✅	Yes
Qwen	✅	✅	Yes
DeepSeek	✅	✅	Yes
Moonshot	✅	✅	Yes
Ollama	✅	✅	No
Claude	❌	❌	Yes
Custom	✅	✅	Depends

Screenshots

After clicking "Test Connection" with valid credentials:

• Success message shows number of available models
• Model input changes from text field to dropdown
• User can select model from list or switch back to manual input

Test plan

• Test connection with OpenAI-compatible endpoint (privnode.com)
• Test connection with Ollama (local)
• Test with invalid credentials (should show error)
• Verify Claude preset shows "not supported" message
• Verify model dropdown appears after successful test
• Verify preset change resets connection state

🤖 Generated with Claude Code

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}