[good first issue] fix(scene): adapt memory prompts to dialogue language

[sirenexcelsior]

Summary

Closes #122.

This PR removes hardcoded Chinese output headings from the scene/persona prompt templates and replaces them with an explicit dialogue-language contract.

What changed

• Added a scene extraction output language contract for scene file names, Markdown headings, and body text.
• Replaced copyable Chinese scene section headings with English fallback headings.
• Added an English-memory rule requiring English memory titles and English section headings.
• Updated persona generation headings to avoid reintroducing Chinese headings into persona.md for non-Chinese scene content.
• Aligned the filename normalizer comments and prompt filename guidance around Unicode letters for multilingual filenames.

Root cause

The scene extraction template included hardcoded Chinese Markdown section headings. LLMs may copy those headings into persisted scene/persona memory files even when the dialogue language is English, German, Russian, French, or another non-Chinese language.

Validation

• Static search for the target Chinese output headings in prompt source files returned no matches.
• npm run build passed.
• Manual LLM smoke tests with deepseek-v4-flash passed for English, German, Russian, and French scene extraction prompts: no Chinese heading leakage and no unintended English fallback headings for non-English cases.

[Maxwell-Code07]

Thank you for participating in our good first issue event, we will complete the review internally as soon as possible!