Add MiniMax speech synthesis endpoint

[octo-patch]

Reason: Add the missing MiniMax text-to-speech route with model selection, regional endpoint configuration, supported request fields, and decoded audio responses.

Changes

• Add POST /v1/audio/speech for synchronous and streamed speech synthesis.
• Support the current speech model set, HTTP request fields, audio formats, and global or China endpoints.
• Decode hexadecimal audio responses into binary output while preserving URL responses as JSON.
• Add mocked coverage for field forwarding, regional routing, response decoding, streaming, and API errors.

Checks

• uv run python -m unittest test_speech.py
• uv run python -m py_compile server.py test_speech.py
• uv run python tests.py --help
• git diff --check