fix(prompt): unify generate_voiceover output protocol to JSON-only

[fancyboi999]

Summary

• unify voiceover system prompts to require plain JSON object output (no markdown/code block)
• align system prompts with existing user prompts in both English and Chinese

Why

Current prompts are internally inconsistent:

• system.md asked for markdown-formatted JSON
• user.md asked for JSON-only output

This inconsistency can increase model output variance and JSON parsing instability.

Changes

• prompts/tasks/generate_voiceover/en/system.md
• prompts/tasks/generate_voiceover/zh/system.md

Scope

• prompt-only change
• no code logic changes
• no docs included