feat: add MiniMax as alternative LLM provider

[octo-patch]

Summary

Adds MiniMax as a first-class LLM provider alongside OpenAI for InternGPT's conversation agent.

Key changes:

• New iGPT/controllers/llm_provider.py — MiniMaxLLM class wrapping MiniMax's OpenAI-compatible chat completion API as a LangChain BaseLLM, plus create_llm() factory for provider-agnostic LLM creation
• Updated ConversationBot.init_agent() to accept provider/api_key parameters
• Updated app.py with a provider dropdown selector (OpenAI / MiniMax) in the login UI
• Auto-detection: if MINIMAX_API_KEY is set (without OPENAI_API_KEY), MiniMax is chosen automatically; or set LLM_PROVIDER=minimax explicitly
• MiniMax-specific handling: temperature clamping (MiniMax requires >0), <think> tag stripping, both old and new openai SDK support
• Models: MiniMax-M2.7, M2.7-highspeed, M2.5, M2.5-highspeed

Files changed (7, +547 additions):

• iGPT/controllers/llm_provider.py — new provider module
• iGPT/controllers/ConversationBot.py — use create_llm() factory
• iGPT/controllers/__init__.py — export new symbols
• app.py — provider selector UI + MiniMax login support
• README.md — usage docs with model table
• tests/test_llm_provider.py — 33 unit tests
• tests/test_llm_provider_integration.py — 3 integration tests

Test plan

• 33 unit tests covering MiniMaxLLM, detect_provider, create_llm factory, constants
• 3 integration tests against live MiniMax API (basic completion, highspeed model, factory)
• Manual testing: select MiniMax from dropdown, enter API key, verify agent conversation works

---

This PR is fully backward-compatible — existing OpenAI users see no behavior change.