feat: add MiniMax as first-class LLM provider

[octo-patch]

Summary

Add MiniMax as a built-in LLM provider alongside OpenAI, Anthropic, and Gemini. MiniMax offers high-performance models (M2.7 with 1M context, M2.5 with 204K context) through an OpenAI-compatible API at https://api.minimax.io/v1.

What's included

• New provider module (libs/ai/src/providers/minimax/):

  • types.rs – MiniMaxConfig with API key + base URL, MINIMAX_API_KEY env auto-detect
  • convert.rs – SDK→OpenAI request conversion with temperature clamping (MiniMax rejects temperature=0.0, clamped to 0.01)
  • stream.rs – SSE streaming via reqwest_eventsource, tool-call state tracking
  • provider.rs – Full Provider trait implementation: generate, stream, list_models, build_headers
  • Static model catalogue: MiniMax-M2.7, MiniMax-M2.7-highspeed, MiniMax-M2.5, MiniMax-M2.5-highspeed

• SDK wiring:

  • InferenceConfig – .minimax() and .minimax_config() builder methods
  • ClientBuilder – MiniMax registration in with_inference_config()
  • ProviderRegistry – Auto-registration from MINIMAX_API_KEY env var
  • ProviderKind – MiniMax variant with "minimax" string conversion

• Shared library:

  • ProviderConfig::MiniMax variant with api_key, api_endpoint, auth fields
  • stakai_adapter.rs – build_inference_config() and build_provider_registry_direct() arms

• CLI integration:

  • stakpak auth login --provider minimax --api-key $MINIMAX_API_KEY
  • Config profile template (generate_minimax_profile())
  • BuiltinProvider::MiniMax enum variant
  • Endpoint cleaning, env-var auth resolution, TUI display info

• Documentation: README updated with MiniMax in provider list, auth login example, BYOK config

Tests

• 23 unit tests: temperature clamping, request conversion, message parsing, tool calls, provider config, error parsing, static models, headers
• 3 integration tests (require MINIMAX_API_KEY): generate, streaming, system message – all passing

Design decisions

• Reuses OpenAI wire types (ChatCompletionRequest, ChatMessage, etc.) to minimize code duplication
• Temperature clamping: 0.0 → 0.01 (MiniMax requires (0.0, 1.0] range)
• Static model list (MiniMax doesn't expose a /v1/models endpoint)
• Follows the same patterns as existing Gemini and Stakpak providers

Test plan

• All 23 MiniMax unit tests pass
• All 3 integration tests pass with MINIMAX_API_KEY
• Full workspace cargo check passes
• Existing test suite unaffected (all 48 unit + 33 doc tests pass)
• Verify CI pipeline passes