feat(tools): add MiniMax speech generation

[octo-patch]

Reason: Add a MiniMax speech backend for the executable text-to-speech tool.

Changes:

• Route supported speech models to the global or China /v1/t2a_v2 endpoint.
• Send voice and audio request options and save hex, URL, or streamed audio responses.
• Resolve the MiniMax provider key and parse data.audio, data.status, and base_resp.status_code.
• Add hermetic tests for both regions, request fields, streaming, downloads, and API errors.

Checks:

• uv run --extra dev ruff format --check raven/agent/tools/media_gen.py raven/config/schema.py tests/test_media_gen_speech_minimax.py
• uv run --extra dev ruff check raven/agent/tools/media_gen.py raven/config/schema.py tests/test_media_gen_speech_minimax.py
• uv run --extra dev pytest -q tests/test_media_gen_speech_minimax.py tests/test_config_raven_sections.py tests/test_config_loader.py tests/test_tool_registry_timeout.py (46 passed)
• uv run --extra dev pytest -q (5899 passed, 43 skipped; one failure in unchanged tests/test_cli_theme.py::test_bold_accent_renders_styled_not_bare)