feat(tool/minimax_tts): add MiniMax text-to-speech tool

[octo-patch]

Reason: The Music & Audio toolkit had no MiniMax text-to-speech tool, so speech could not be synthesized with the MiniMax speech models.

Changes

• Add tools/music_audio/minimax_tts/ with tool.json, settings.sample.json, a Node.js tool and a Python tool that implement the same logic.
• Register minimax_tts in tools/music_audio/toolkit.json so the toolkit registry picks it up.

Tool behaviour

synthesizeToFile(text, outputPath, options) (synthesize_to_file in Python) posts to the text-to-audio endpoint and writes the generated audio to outputPath:

• Regional endpoints: global_en calls https://api.minimax.io/v1/t2a_v2 and cn_zh calls https://api.minimaxi.com/v1/t2a_v2. The region comes from the settings and can be overridden per call.
• Models: speech-2.8-hd, speech-2.8-turbo, speech-2.6-hd, speech-2.6-turbo, speech-02-hd, speech-02-turbo, speech-01-hd and speech-01-turbo, defaulting to speech-2.8-hd.
• Authorization: the API key is sent as a Bearer token and never logged.
• Request fields: model and text are always sent, stream is forced to false because the whole audio is needed to write a file, and output_format, language_boost, voice_setting, pronunciation_dict, audio_setting, voice_modify and subtitle_enable are forwarded only when provided. audio_setting.format accepts mp3, wav, flac and pcm.
• Audio handling: the hexadecimal payload returned in data.audio is decoded to bytes, and when output_format is url the audio is downloaded instead.
• Error handling: a non-zero base_resp.status_code is reported with the returned message, a null data object and a malformed payload raise an explicit error instead of writing a truncated file, and the model, region, audio format and output path are validated before any request is sent.

Settings

MINIMAX_TTS_API_KEY is required, so the tool is reported as unavailable until it is filled. MINIMAX_TTS_MODEL and MINIMAX_TTS_REGION default to speech-2.8-hd and global_en.

Checks

• tsc --noEmit -p tools/tsconfig.json reports no error in the added files.
• python3 -m py_compile on the added Python files.
• Local verification, not part of the diff: both implementations were exercised against a stubbed network layer, asserting the resolved host and path per region, the Bearer header, the exact request body for a full and a minimal call, every supported model, the hex decoding and the url download, and each validation and error path. The manifest was also checked against the registry expectations: the tool is discovered through toolkit.json, the runtime resolves src/nodejs/index.ts, MINIMAX_TTS_API_KEY is the only required setting, and the declared parameters map onto the method signature with the optional argument trailing.