Add MiniMax as third AI provider

[octo-patch]

Summary

Adds MiniMax as a third AI provider alongside Anthropic and OpenAI. MiniMax exposes an OpenAI-compatible API with models like M2.7 (1M context window), M2.5, and M2.5-highspeed (204K context, fastest).

Changes

• lib/minimax-auth.ts — Auth resolution with priority: override key → DB key → MINIMAX_API_KEY env var → proxy via MINIMAX_BASE_URL
• lib/ai-client.ts — MiniMaxAIClient class using OpenAI SDK with thinking-tag stripping for M2.5+ model responses
• lib/settings.ts — MiniMax model cache + getProvider() now returns 'minimax' as a valid provider
• app/api/settings/route.ts — MiniMax model validation (MiniMax-M2.7, MiniMax-M2.5, MiniMax-M2.5-highspeed), API key CRUD
• app/api/settings/test/route.ts — MiniMax connection test endpoint
• app/settings/page.tsx — 3-way provider toggle (Anthropic / OpenAI / MiniMax) with model selector and key management
• .env.example — MINIMAX_API_KEY and MINIMAX_BASE_URL documentation
• README.md — MiniMax in tech stack table and configuration docs

Why MiniMax?

• 1M context window (M2.7) — ideal for processing large batches of bookmarks in fewer API calls
• OpenAI-compatible API — minimal integration effort, reuses the existing OpenAI SDK
• Competitive pricing — good alternative for users who want to try different providers

Test plan

• 30 tests added (12 auth unit, 14 client unit, 4 integration) — all passing
• TypeScript compiles cleanly (npx tsc --noEmit)
• Manual: select MiniMax provider in Settings, add API key, test connection
• Manual: run categorization pipeline with MiniMax provider selected
• Manual: verify thinking tags are stripped from M2.5 responses

13 files changed, 799 additions(+), 22 deletions(-)