feat: add MiniMax Cloud API as alternative prompt rewrite provider

[octo-patch]

Summary

• Add MiniMax Cloud API as an alternative prompt rewrite provider for text-to-video (T2V) generation
• Users can switch to MiniMax by setting REWRITE_PROVIDER=minimax and MINIMAX_API_KEY, eliminating the need to deploy a local vLLM server
• Support MiniMax-M2.7 (latest), MiniMax-M2.5, and MiniMax-M2.5-highspeed (204K context) models

Changes

• hyvideo/utils/rewrite/clients.py: Add MiniMaxClient class with OpenAI-compatible API calls, temperature clamping to [0.0, 1.0], and <think> tag stripping
• hyvideo/utils/rewrite/rewrite_utils.py: Add _create_t2v_client() factory with REWRITE_PROVIDER env var for provider selection (defaults to existing Qwen/vLLM behavior)
• README.md / README_CN.md: Document MiniMax configuration with env vars and available models
• tests/test_minimax_client.py: 24 unit tests + 3 integration tests covering init, temperature clamping, think-tag parsing, provider selection, and end-to-end rewrite

Usage

export REWRITE_PROVIDER=minimax
export MINIMAX_API_KEY="your_api_key"
# Optional: override model (default: MiniMax-M2.7)
# export T2V_REWRITE_MODEL_NAME="MiniMax-M2.5-highspeed"

Test plan

• 24 unit tests pass (mock-based, no API key required)
• 3 integration tests pass with real MiniMax API
• Existing Qwen/vLLM provider behavior unchanged when REWRITE_PROVIDER is not set
• Verify prompt rewriting quality with actual video generation pipeline