feat: add MiniMax as direct LLM provider for podcast generation

[octo-patch]

Summary

Adds MiniMax M2.7 as a first-class LLM provider alongside OpenRouter and xAI. Since Twocast already uses MiniMax for TTS, users can now reuse the same MINIMAX_TOKEN for LLM by simply setting LLM_PROVIDER=minimax.

What changed

• New src/utils/llm.ts: Provider preset system with auto-detection, temperature clamping, and think-tag stripping for M2.7
• Updated src/utils/xai.ts: Uses provider config instead of raw env vars; fully backward-compatible
• Updated .env.example: Documented LLM_PROVIDER and MINIMAX_API_KEY
• Updated README.md and README.zh-CN.md: Provider comparison table with MiniMax docs
• 31 unit tests + 3 integration tests (all passing)

Why MiniMax?

• MiniMax M2.7 offers a 1M token context window, ideal for long documents and podcast scripts
• OpenAI-compatible API integrates seamlessly
• Users with MINIMAX_TOKEN for TTS get LLM support with zero additional setup

Test plan

• 31 unit tests covering provider detection, config resolution, temperature clamping, think-tag stripping
• 3 integration tests verifying full request pipeline
• Manual test: set LLM_PROVIDER=minimax and generate a podcast
• Verify backward compatibility: existing OpenRouter config still works

7 files changed, 557 additions(+), 27 deletions(-)