feat: add MiniMax as LLM provider for classification/summarization

[octo-patch]

Summary

Adds MiniMax as a first-class LLM provider for memory type classification and auto-summarization, alongside the existing OpenAI support.

What changed

• LLM_PROVIDER env var (auto/openai/minimax) for explicit provider selection
• Auto-detection: when OPENAI_API_KEY is not set, falls back to MINIMAX_API_KEY automatically
• Temperature clamping to (0, 1] for MiniMax models (API requirement)
• think tag stripping for MiniMax M2.5+ reasoning traces in classification/summarization responses
• MiniMax-M2.7 (204K context) as default model when using MiniMax
• Documentation updates: .env.example, README.md, INSTALLATION.md, CLAUDE.md, docs/ENVIRONMENT_VARIABLES.md

Why MiniMax?

MiniMax M2.7 offers a 204K context window at competitive pricing (~$0.14/1M input tokens), making it a cost-effective alternative for memory classification and summarization tasks. Its OpenAI-compatible API means zero new dependencies - it reuses the existing openai Python SDK with a different base_url.

Quick start

# Option 1: Explicit provider
LLM_PROVIDER=minimax
MINIMAX_API_KEY=your-key
CLASSIFICATION_MODEL=MiniMax-M2.7

# Option 2: Auto-detection (just set the key)
MINIMAX_API_KEY=your-key

Files changed

File	Change
automem/config.py	Add LLM_PROVIDER, MINIMAX_API_KEY, MINIMAX_BASE_URL, MINIMAX_DEFAULT_MODEL
automem/service_runtime.py	Extend init_openai() with MiniMax auto-detection and explicit selection
automem/classification/memory_classifier.py	Temperature clamping + think-tag stripping for MiniMax models
automem/utils/text.py	Temperature clamping + think-tag stripping for MiniMax in summarization
.env.example	MiniMax config section
README.md	Add MiniMax to Configuration section
INSTALLATION.md	Add MiniMax env vars to config table
CLAUDE.md	Add LLM_PROVIDER and MINIMAX_API_KEY docs
docs/ENVIRONMENT_VARIABLES.md	Add MiniMax model pricing + LLM provider docs
tests/test_minimax_provider.py	26 unit tests covering init, temp clamping, think-tag stripping, summarization
tests/test_minimax_integration.py	4 integration tests (classification + summarization with real API)

11 files changed, 840 additions, 16 deletions

Test plan

• 26 unit tests pass (pytest tests/test_minimax_provider.py)
• 4 integration tests pass (MINIMAX_API_KEY=... pytest tests/test_minimax_integration.py)
• Existing test_service_runtime.py tests still pass (backward compatible)
• Verify LLM_PROVIDER=minimax with real MiniMax API key
• Verify LLM_PROVIDER=auto fallback from OpenAI to MiniMax
• Verify existing OpenAI workflow is unaffected