feat: add MiniMax as first-class LLM provider with M3 default

[octo-patch]

Summary

• Add MiniMax as a first-class LLM provider alongside OpenAI and LMStudio
• Ship MiniMax-M3 as the default model (512K context window, 128K max output, image input support)
• Also expose MiniMax-M2.7 and MiniMax-M2.7-highspeed as alternatives
• Auto-detect MiniMax via MINIMAX_API_KEY / CC_MINIMAX_API_KEY environment variables
• Return static model list for MiniMax (no /v1/models endpoint available)
• Verify MiniMax API key via chat completions endpoint

Why M3

MiniMax-M3 is the latest MiniMax model. Compared with M2.7 it ships a larger 512K context window, up to 128K output, and image input support on the OpenAI-compatible endpoint — making it the right default for new ComfyUI-Copilot users while keeping M2.7 / M2.7-highspeed available for backwards compatibility.

Model list (final)

Model ID	Notes
MiniMax-M3	512K context, 128K max output, image input — default
MiniMax-M2.7	Previous-generation model
MiniMax-M2.7-highspeed	Previous-generation low-latency variant

Older models (M2.5 / M2.1 / M2 / M1) are intentionally not included.

Changes

• backend/utils/globals.py: is_minimax_url(), MINIMAX_MODELS (M3 first), env var support, auto-detect in apply_llm_env_defaults()
• backend/agent_factory.py: Default to MiniMax-M3 when base_url points to MiniMax API
• backend/controller/llm_api.py: Static MiniMax model list (M3 first), key verification via chat completions using M3
• README.md, README_CN.md: MiniMax configuration documentation (env vars + UI) describing M3 (512K / 128K) + M2.7 alternatives
• tests/: Unit tests + integration tests with ComfyUI runtime stubs, including M3-default / M3-first / older-models-removed coverage

Test plan

• Unit tests pass (URL detection, MINIMAX_MODELS contents including M3 first / older models removed, env defaults, agent factory default = M3)
• Integration tests updated to use MiniMax-M3 against the real MiniMax API
• Manual smoke test in ComfyUI with a MiniMax API key

[CLAassistant]

All committers have signed the CLA.

[Jefsky]

Hi @octo-patch! Just a heads-up: the base_url https://api.minimax.io/v1 was tested in a related Pixelle-Video project and couldn't fetch the model list. The working URL is https://api.minimax.chat/v1 — that one successfully returns the model list (e.g. MiniMax-M2.7). You may want to update this PR to use the correct URL. Thanks!