feat: add MiniMax as first-class LLM provider

[octo-patch]

Summary

• Add MiniMax cloud models (MiniMax-M2.7, MiniMax-M2.5, MiniMax-M2.5-highspeed) as a dedicated LLM backend via --llm minimax
• Auto-detects MINIMAX_API_KEY from environment, clamps temperature to [0.0, 1.0], strips <think> reasoning tags from M2.5/M2.7 responses
• Adds MiniMaxModelHandler extending OpenApiModelHandler, dedicated arguments dataclass, pipeline registration, README docs

Changes

File	Description
LLM/minimax_language_model.py	MiniMax handler with temp clamping and think-tag stripping
arguments_classes/minimax_language_model_arguments.py	Dedicated arguments dataclass with MiniMax defaults
s2s_pipeline.py	Register minimax as LLM option in pipeline
arguments_classes/module_arguments.py	Update --llm help text
README.md	Add MiniMax to LLM list and usage section
tests/test_minimax_language_model.py	21 unit + 3 integration tests

Usage

export MINIMAX_API_KEY="your-api-key"

python s2s_pipeline.py \
    --llm minimax \
    --minimax_model_name MiniMax-M2.7 \
    --minimax_stream \
    --tts pocket \
    --recv_host 0.0.0.0 \
    --send_host 0.0.0.0

Test plan

• 18 unit tests pass (think-tag stripping, temperature clamping, API key detection, defaults, argument renaming)
• 3 pipeline integration tests pass (subclass check, argument renaming, process override)
• 3 integration tests with real MiniMax API (non-streaming, streaming, session reset)
• Existing test_openai_api_language_model.py tests still pass (no regressions)