fix: increase ingest timeout to 300s and cap images at 15 per listing ...

- Prevents inline pipeline timeouts on listings with 40+ photos
- 15 images is enough for AI analysis while keeping request duration reasonable

feat: download images locally and ingest via multipart /with-images endpoint ...

- Update collector to support ingest_with_images using CurlMime multipart
- Crawler now downloads listing photos in parallel (5 workers) to temp files
- Fix photo URL builder: use photosnew CDN + beautifulUrl (legacy file paths are 415)
- Add fallback title generation in normalizer for PayloadSchema validation
- Increase ingest timeout to 120s for inline pipeline processing

Closes: photo ingestion via binary upload

Implement MVP DOM.RIA parser with curl_cffi scraping ...

- Add src/ modules: config, session, parser, normalizer, collector, storage, crawler, main
- Implement curl_cffi session with chrome124 impersonation and cookie warmup
- Extract __INITIAL_STATE__ using bracket-counting JSON parser
- Normalize DOM.RIA data into data_collector payload format
- Add SQLite resume storage for deduplication
- Implement rate limiting: 10s between pages/details, 2min pause every 50 pages
- Add CLI entry point with argparse for city/category/operation selection
- Add Dockerfile and requirements.txt
- Update TECH_SPEC.md with v3.0 post-MVP status and empirical data findings
- Update README.md with usage instructions

Co-Authored-By: Claude <noreply@anthropic.com>