fix(upload): stop large photo batches failing as "network error"

[Shaan-alpha]

Problem

Uploading a real photo gallery failed with a red "Network error — is the server running?" toast right after clicking Start Sorting — even though the server was fine.

Root cause: a real gallery exceeds Werkzeug's 1000-part multipart limit and/or the byte cap. The dev server rejects an over-limit request before reading the body, so the browser (still uploading) sees the socket reset and fetch() throws — surfacing as a misleading "network error" instead of a clean message. Raising the byte cap alone (a prior fix) was whack-a-mole and never touched the separate 1000-part ceiling.

Fix

• config — add MAX_UPLOAD_FILES (5000 local / 500 public, env-tunable via FACE_SORT_MAX_UPLOAD_FILES) and raise MAX_FORM_PARTS just above it so a valid batch is never reset; expose both limits to the page.
• frontend — pre-flight the batch size/count before uploading and show a clear, actionable message (with a Local Folder nudge for big galleries) instead of letting the browser reset; parse the response defensively so an HTML error body isn't mislabeled as a network error.
• tests — cover the new limits, env override, public-mode tightening, and acceptance of a batch over the historical 1000-part ceiling.

Verification

• pytest tests/ → 21 passed (6 new)
• Manually verified end-to-end from source: uploads, SSE progress, results, and zip downloads all succeed.
• node --check clean on app.js; template renders the injected limits.

🤖 Generated with Claude Code