Add MiniMax Text to Speech skill

[Octopus (octo-patch)]

Reason: Add a local MiniMax text-to-speech skill implementing the global/CN /v1/t2a_v2 and /v1/t2a_async_v2 endpoints, speech-2.8-hd family models, request fields, audio formats, and async/WebSocket query operations.

Changes

• Adds minimax-text-to-speech/SKILL.md documenting synchronous HTTP, asynchronous, and realtime WebSocket synthesis workflows for the MiniMax text-to-speech API.
• Adds minimax-text-to-speech/scripts/text_to_speech.py, a standard-library-only Python CLI with http, async, query, and ws subcommands:
  • POST /v1/t2a_v2 synchronous synthesis with the speech-2.8-hd model family, the documented request fields (model, text, stream, language_boost, output_format, voice_setting, pronunciation_dict, audio_setting, voice_modify, subtitle_enable), and mp3/wav/flac/pcm output.
  • POST /v1/t2a_async_v2 async task creation plus /v1/query/t2a_async_query_v2 status polling, with GET /v1/files/retrieve download of the finished audio file.
  • WSS /ws/v1/t2a_v2 realtime synthesis using a standard-library WebSocket client.
  • Global (api.minimax.io) and China (api.minimaxi.com) regions.
• Adds a Creative & Media README entry linking to the new skill.

Checks

• python3 -m py_compile minimax-text-to-speech/scripts/text_to_speech.py
• CLI --help for all four subcommands
• Unit checks for request payload building, hex/base64 audio decoding, single-JSON and SSE response parsing, and WebSocket frame round-trips
• End-to-end mock-server checks for the http, async create, query, and file retrieve flows
• Secret scan clean