[codex] Improve XHS media text pipeline

[jian45154]

Summary

• Extract Xiaohongshu subtitle tracks, transcripts, carousel images, shortlink note IDs, and canonical URLs more reliably.
• Wire video transcripts into media text ingestion/reindexing and improve media pipeline status accounting.
• Make the Tesseract OCR preset usable on Windows, validate tessdata languages, handle empty OCR as no_text, and preprocess animated GIFs through ffmpeg first frames.
• Add regression coverage for Xiaohongshu shortlinks, subtitles, carousel media, OCR presets, no_text, GIF preprocessing, and media pipeline blockers.

Why

The live URL batches exposed missing pieces in the media pipeline: video transcripts were not fully usable as media text, image OCR was blocked by runtime discovery/language data, empty OCR output kept items stuck in not_processed, and animated GIFs failed Tesseract directly.

Validation

• python -m pytest -> 249 passed
• git diff --check
• Live shortlink batch: 5/5 succeeded, media pipeline ready
• Mixed WeChat/XHS batch: media pipeline ready after OCR/no_text/GIF handling