Rebuild Vinted side on its own JSON API - fixes both remaining bugs

Root-caused why the avatar filter and item count were still wrong for
a real, larger, logged-in closet (not reproducible with my anonymous
14/39-item test accounts):

- Fresh testing showed Vinted's grid genuinely does not scroll-load
  more items via a simple scrollTo(bottom) - confirmed by scrolling
  and waiting up to 18s on a 96-result catalog page with the count
  never moving past the initial batch. The DOM-scraping approach was
  fundamentally guessing at a lazy-load mechanism that doesn't work
  the way I'd assumed, which explains the undercounts.
- Inspecting the page's own network requests found the real
  mechanism: a clean, same-origin JSON API the page itself calls,
  /api/v2/wardrobe/<user_id>/items?page=N&per_page=50, returning
  title/brand/size/status/price/photos already structured - and the
  photos are item-scoped, so there's no avatar to filter out at all
  (that was a separate field entirely: item.user.photo).

Rebuilt the Vinted bookmarklet around this API: page through it until
a short page confirms the end (reliable, complete count - verified
41/41 items on a real closet, vs. 20 in the initial DOM-scraped
render), then still fetch each item's own page once for the fuller
free-text description via the same ld+json block as before, since the
wardrobe API only returns the short title. If that per-item fetch
fails, the draft now falls back to the title alone instead of being
dropped - the wardrobe data already covers everything else.

Depop's side is unchanged - it has no equivalent API (confirmed
earlier: blocked by Cloudflare bot management server-side, no CORS
browser-side), so DOM-scraping remains the only option there.

Verified against a real, live, larger closet (41 items, not
reproducible with the smaller test accounts used earlier) - full
pagination, correct brand/size/condition/price/photos, zero avatar
contamination by construction rather than by filtering.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>