feat: add MiniMax TTS backend

[octo-patch]

Reason: The existing voice pipeline has no MiniMax TTS backend for the global and CN speech endpoints.

This change:

• adds global and CN MiniMax TTS endpoint selection with the current speech model presets
• maps voice text requests to PCM output and validates API response errors and hex audio
• routes prompt and streaming voice flows through the configured TTS backend
• adds focused tests for backend selection, request mapping, responses, and streamed text collection

Checks:

• uv run --python 3.11 --no-project --with-requirements backend/requirements.txt pytest -q backend/tests/test_unit_minimax_tts.py
• python3.11 -m py_compile backend/core/voice_service.py backend/tests/test_unit_minimax_tts.py
• git diff --check