feat(tts): extend Doubao controls and caption rendering

[RayJiang4S]

Summary

This builds on the merged Doubao Speech provider by adding the controls we needed while producing longer Mandarin narration:

• expose advanced Doubao request parameters such as model, emotion, emotion_scale, loudness_rate, mute-cut controls, pitch post-processing, and raw escape hatches for newly documented fields
• document a production workflow for reducing theatrical delivery without rewriting approved narration
• add caption overlay sizing options for Remotion explainers
• preserve newlines in text cards and avoid fully transparent entrance frames
• align Remotion metadata duration with rounded sequence frame math so paused playback lands on the final visual frame

Verification

• python3 -m py_compile tools/audio/doubao_tts.py
• git diff --check c2f83f1^ c2f83f1