Add chapter context translation mode

[DygDyg]

Summary

This PR adds an optional Chapter Context Translation mode for LLM-based translation.

Instead of translating each page independently, Koharu can now process translation using shared chapter-level context.

Pipeline flow in chapter mode:

1. Detection across selected pages
2. OCR across selected pages
3. Collect OCR blocks with stable block IDs
4. Chunk blocks into token-limited LLM requests
5. Translate with validation + retry/fallback
6. Apply translations back to original nodes
7. Continue inpainting and rendering

The default page-by-page translation workflow remains unchanged.

---

Features

Chapter Context Translation

• Added optional chapter-wide translation mode
• Stable block IDs (P001-B001, etc.)
• Translation preserves reading order across pages

Chunking

• Token-based chunking
• Configurable token budget
• Configurable max blocks per chunk

Reliability Improvements

• Output validation for missing/extra blocks
• Detection of repetitive malformed output
• Retry with stricter prompt
• Single-block fallback for failed chunks
• All-or-nothing apply preserved

UI

• Settings toggle for Chapter Context Translation
• Advanced chunk configuration
• Progress bubble hint when enabled
• OCR overlay background customization (RGBA)

Localization

• Added Russian localization for new UI strings

---

Notes

• Standard translation mode is unchanged.
• Chapter context mode only applies to full pipeline runs.
• Per-page translation and single-block translation still use the original path.

Tested with:

• Local LLM providers
• OpenAI-compatible providers
• Chapter translation chunk validation
• UI settings and pipeline hint behavior

[github-actions]

Thanks for your first PR to Koharu.

Please review our contribution guide before review:
https://koharu.rs/contribute/introduction/

In the PR description, include:

• what changed
• any user-visible behavior differences
• how you verified the change

If AI helped produce the patch, a human still needs to review and understand it before submission.

[DygDyg]

Notes for maintainer

This PR became larger than initially planned because several related improvements were needed for stable chapter-context translation.

If preferred, I can split this into multiple smaller PRs:

• core chapter translation
• UI improvements
• Windows CUDA build fixes