feat: add MiniMax as first-class LLM provider

[octo-patch]

Summary

Add MiniMax as a first-class LLM provider in ChainForge, enabling users to query MiniMax M2.7 and M2.7-highspeed models directly from the visual prompt-testing UI.

Changes

• Provider registration (models.ts): NativeLLM enum entries, LLMProvider.MiniMax, getProvider() detection, rate limit (1000 RPM)
• API integration (utils.ts): call_minimax() function using OpenAI-compatible API at https://api.minimax.io/v1, temperature clamping (min 0.01) per MiniMax requirement, response extraction via existing OpenAI handler
• Bugfix (utils.ts): call_chatgpt() now respects custom API_KEY parameter - previously it threw no OpenAI key even when a custom key was passed by providers like DeepSeek/MiniMax
• Settings schema (ModelSettingSchemas.tsx): Full model settings form with temperature (0.01-1.0, default 0.7), system message, top_p, max_tokens, stop sequences, presence/frequency penalty
• UI menu (store.tsx): MiniMax group with M2.7 and M2.7-highspeed entries
• Backend env mapping (flask_app.py): MINIMAX_API_KEY environment variable support
• Tests (minimax.test.ts): 25 tests - 21 unit tests (enum values, provider detection, schema validation, response extraction, form data processing) + 3 integration tests (live API call, temperature clamping, system message) + 1 conditional test

How it works

MiniMax follows the same pattern as DeepSeek: an OpenAI-compatible wrapper that delegates to call_chatgpt() with a custom base URL and API key. Users just need to set their MINIMAX_API_KEY in Settings.

Test plan

• All 21 unit tests pass
• All 3 integration tests pass (with MINIMAX_API_KEY set)
• No regressions in existing test suite (pre-existing failures remain unchanged)
• Manual verification: add MiniMax model from UI dropdown, query with prompt, verify response