Add PDF extraction, lazy scroll, and image OCR to crawler

fix: set HOME=/app so fastembed cache is writable by user 1000

Fix Playwright browser path for non-root user on HF Spaces

fix: playwright missing + crawler/loader embedding model mismatch

perf: pre-download fastembed ONNX model at build time

fix: remove invalid --no-cache flag from uv pip install

fix: remove --prefer-binary (breaks packages), keep cmake for C++ builds

fix: --prefer-binary + cmake to prevent C++ source compilation on HF

fix: install uv via pip instead of curl (curl was hanging on HF)

Switch to uv for fast package installs (10-100x faster than pip)

Fix pip install hanging: prefer-binary + unpin lxml

bust docker cache for fresh build

add HF Spaces Docker deployment