Carousels return a mixed image+video URL list. The uploader keyed on the
post's media_type, so every carousel URL went through the image optimizer;
videos failed PIL ('cannot identify image file') and were then re-downloaded
for the raw fallback. That double download per video pushed the synchronous
scrape past the backend's request timeout, so the backend imported 0 while
the engine still saved posts to its own DB (causing a dedup desync on retry).
Detect video URLs up front and send them straight to the video uploader.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Stop resuming from the persisted (older) BoxAPI cursor. Resuming made every
sync walk further back in time and never re-fetch posts published since the
last run, so a seller's newest posts were never imported. Always start from
max_id=None (newest) and page older within a single run, de-duping against
the DB: brand-new posts first, then older un-imported ones.
Also emit timezone-aware UTC created_at so ordering is correct regardless of
the container's local timezone (and Django no longer receives naive datetimes).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
