feat: add MiniMax as alternative LLM provider

[octo-patch]

Summary

Add MiniMax (MiniMax-M2.7, MiniMax-M2.7-highspeed with 204K context) as a first-class LLM provider for 4KAgent reasoning agent, alongside the existing GPT-4 and Azure OpenAI backends.

Changes

• New llm/minimax.py: MiniMaxLLM class extending BaseLLM, using MiniMax OpenAI-compatible API at api.minimax.io/v1, with:
  • Temperature clamping to (0.0, 1.0] (MiniMax API constraint)
  • Think-tag stripping for M2.7 models
  • Exponential backoff retry with rate-limit detection
• config.yml: Added MINIMAX section with API_KEY, MODEL, MAX_TOKENS, TEMPERATURE
• llm/__init__.py: Export MiniMaxLLM
• pipeline/the4kagent_pipeline.py: Added llm_provider parameter - supports gpt4 (default), azure, and minimax
• infer_4kagent.py: Added --llm_provider CLI argument with choices [gpt4, azure, minimax]
• README.md: Documented MiniMax provider setup and usage with example command
• tests/: 38 unit tests + 6 integration tests (44 total, all passing)

Usage

# Set MINIMAX API key in config.yml, then:
python infer_4kagent.py \
  --input_dir ./dataset/LQ \
  --output_dir ./outputs/results \
  --profile_name ExpSR_s4_F \
  --llm_provider minimax

Why MiniMax?

• 204K context window - handles complex multi-step restoration plans
• OpenAI-compatible API - drop-in replacement, no SDK changes needed
• Cost-effective - competitive pricing for research workloads

Test plan

• 38 unit tests covering init, config, request payload, response validation, think-tag stripping, retry logic, format checking
• 6 integration tests against live MiniMax API (simple query, JSON format check, system message, model variants, pipeline signature)
• All 44 tests passing
• Backward compatible - existing GPT-4 and Azure workflows unchanged (default --llm_provider gpt4)