feat(tools): add music_generate tool via MiniMax

[octo-patch]

Reason: Add a music_generate tool backed by the MiniMax /music_generation endpoint for global and China regions.

Changes

• Add MusicGenerateTool to the media tool registry (raven/agent/tools/media_gen.py): one synchronous POST to {api_base}/music_generation, with url and hex output formats, cover reference audio (audio_url / audio_base64), streamed hex chunks, and the optional aigc_watermark field required on the China endpoint.
• Default model is music-3.0; the model parameter documents the generation models (music-3.0, music-2.6, music-3.0-free, music-2.6-free) and cover models (music-cover, music-cover-free).
• Response parsing follows base_resp.status_code (0 = success), data.status (2 = ready), and data.audio (hex payload or downloadable URL list).
• Add tools.media.music to MediaGenConfig and resolve its API key from the minimax provider config as a fallback in effective_media_config.
• Register the tool in AgentLoop._register_default_tools when tools.media.music has a key or model.
• Add hermetic unit tests (tests/test_media_gen_music.py) covering url/hex output, the global and China endpoints, cover reference audio, streamed hex, and error parsing.

Checks

• pytest tests/test_media_gen_music.py tests/test_tool_registry_timeout.py tests/test_tool_search.py -> 47 passed
• ruff check on the changed files -> All checks passed
• ruff format --check on the changed files -> 5 files already formatted