feat: add MiniMax provider support with M3 as default

Summary

This PR adds MiniMax chat model support to @open-harness/agent via the Vercel AI Gateway, with MiniMax-M3 as the default model.

Changes

• packages/agent/models.ts: Add shouldApplyMiniMaxDefaults() to detect minimax/* model IDs, and apply a default temperature: 1 via defaultSettingsMiddleware. MiniMax's API rejects temperature === 0, so this default ensures callers that omit temperature get a valid value. Comment updated to call out M3 (default), M2.7, and M2.7-highspeed coverage.
• packages/agent/index.ts: Export shouldApplyMiniMaxDefaults from the package index.
• packages/agent/models.test.ts: Add unit tests for shouldApplyMiniMaxDefaults and MiniMax getProviderOptionsForModel, covering M3 (default), M2.7, and M2.7-highspeed.

Supported models

All three IDs are already part of GatewayModelId in @ai-sdk/gateway and are routed through the existing Vercel AI Gateway. Older variants (M2.5/M2.1/M2/M1) are not included.

Why M3

MiniMax-M3 is the latest MiniMax model. Highlights:

• 512K context window
• Up to 128K max output tokens
• Image input supported (text + images; video/audio/documents not supported)
• OpenAI-compatible and Anthropic-compatible APIs

API references

• Chat (OpenAI Compatible): https://platform.minimax.io/docs/api-reference/text-openai-api
• Chat (Anthropic Compatible): https://platform.minimax.io/docs/api-reference/text-anthropic-api
• Pricing: https://platform.minimax.io/docs/guides/pricing-paygo

Test plan

• Unit tests for shouldApplyMiniMaxDefaults (M3 + M2.7 + M2.7-highspeed) pass
• Unit tests for getProviderOptionsForModel with MiniMax model IDs pass
• bun test packages/agent/models.test.ts — 19 pass, 0 fail