Add MiniMax text-to-speech API tool for global and CN endpoints

[octo-patch]

Reason: Add a MiniMax text-to-speech API tool exposing both the global (api.minimax.io) and China (api.minimaxi.com) T2A endpoints.

Changes:

• api.py: add minimax_tts_api(prompt, ...), which posts to the MiniMax /v1/t2a_v2 text-to-audio endpoint. It selects the regional base URL via the region argument (global_en or cn_zh), sends model, text, stream, and output_format, decodes the hex audio payload returned in data.audio, verifies base_resp.status_code == 0, and writes the audio file to disk. The supported models and audio formats are listed as module constants.
• phi_3_vision_mlx.py: re-export minimax_tts_api from api so it is available to the Agent toolchain.
• gte.py: add a MiniMax T2A code snippet to the API retrieval list used by get_api, alongside the existing text-to-speech entry.
• examples.py: add Agent toolchain examples exercising the global endpoint, the China endpoint, and an alternate MiniMax TTS model.

Checks:

• python3 -m py_compile api.py gte.py examples.py (the shared environment runs Python 3.10; the full MLX test suite requires Python 3.12 and Apple silicon, so only the changed, import-light modules were syntax-checked here).

Target parameters (model list, default model, audio formats, regional URLs, required request fields, and the data.audio / base_resp.status_code response shape) are sourced from the task's multimodal speech configuration.

[JosefAlbers]

Thanks for the contribution! This looks reasonable, but I can't validate it locally without API keys. Could you run the change on your end and attach the command/output (or a screenshot) showing it works? Once there's evidence it's been tested, I'll take another look.