feat: Add dynamic model discovery from provider APIs

[TerminallyLazy]

Replace static model name text inputs with dynamic dropdowns that fetch
available models directly from each provider's API based on configured
API keys.

Supported providers:

• OpenAI, Anthropic, Google Gemini, Groq, Mistral, DeepSeek, xAI
• OpenRouter, SambaNova
• Any OpenAI-compatible provider with custom api_base

Changes:

• Add python/helpers/model_discovery.py for API-based model fetching
• Add python/api/settings_refresh_models.py endpoint
• Convert model name fields from text to select with dynamic options
• Add frontend handlers for provider change and custom model entry
• Include 1-hour cache with force refresh capability
• Preserve "Custom (enter manually)" option for unlisted models

Backend

• python/helpers/model_discovery.py: Core module that fetches models from each provider's /models endpoint with provider-specific authentication (Bearer token, Anthropic headers, Google query params)
• python/api/settings_refresh_models.py: API endpoint called by frontend, resolves masked API key placeholders to actual values from environment

Frontend

• handleSelectChange(): Triggers model refresh when provider dropdown changes
• refreshAllModelOptions(): Fetches all model lists when settings modal opens
• Custom model input: Falls back to text input when "Custom (enter manually)" is selected
• Alpine.js reactivity fix: Uses splice() for in-place array modification to ensure UI updates

Caching

• File-based cache at tmp/model_cache.json
• 1-hour TTL with per-provider, per-type granularity
• Force refresh option bypasses cache on provider change
• Clear cache option available for troubleshooting