Add MiniMax as alternative LLM provider

[octo-patch]

Summary

Adds MiniMax as an alternative LLM provider alongside OpenAI, configurable via environment variables.

Changes

• New llm_provider.py — provider abstraction module with auto-detection (MINIMAX_API_KEY to MiniMax), temperature clamping (MiniMax requires (0, 1]), <think> tag stripping for reasoning models, and openai v0.x / v1.x SDK compatibility
• Updated access.py — replaced hardcoded openai.ChatCompletion.create() + key cycling with the provider module; legacy tokens.txt key rotation preserved as fallback
• README — added LLM Provider Configuration section with MiniMax setup instructions and available models
• 35 tests — 31 unit tests (test_llm_provider.py) + 4 integration tests (test_integration_minimax.py), all passing

Configuration

# Use MiniMax (auto-detected if MINIMAX_API_KEY is set)
export MINIMAX_API_KEY=your-key

# Or explicit
export LLM_PROVIDER=minimax
export LLM_MODEL=MiniMax-M2.7  # default, 204K context

Available MiniMax models: MiniMax-M2.7, MiniMax-M2.7-highspeed, MiniMax-M2.5, MiniMax-M2.5-highspeed

Files changed (5 files, 670 additions)

File	Change
tools/back-end/llm_provider.py	New provider abstraction module
tools/back-end/access.py	Use provider module instead of hardcoded OpenAI
tools/tests/test_llm_provider.py	31 unit tests
tools/tests/test_integration_minimax.py	4 integration tests
README.md	Provider config docs

Test plan

• All 35 tests pass
• Existing OpenAI flow unchanged (default provider)
• MiniMax provider resolves correct API base, model, and key
• Temperature clamping prevents invalid values for MiniMax
• Think-tag stripping works for MiniMax reasoning models
• Legacy tokens.txt key rotation still works
• Live MiniMax API integration test passes