feat(upload): skip unreadable files (FIFOs/sockets) + auto-reload on new deploy

Two robustness fixes behind the recurring "folder upload stuck at ~93%" reports.

1. Skip non-regular files up front. A copied s6/runit service tree contains
   FIFOs (e.g. supervise/control named pipes) that report a size but BLOCK
   FOREVER when the browser reads them — the deterministic ~8-files-short that
   no retry/watchdog tweak could fix. uploadBatch/uploadTree now probe each
   file's first chunk against a 3 s timeout (partitionReadable), upload only the
   readable ones, and report the rest: "N uploaded · M skipped (not regular
   files)". Progress runs over the uploadable count, so it reaches 100% instead
   of parking at 93% while a lane hangs on a pipe.

2. Auto-reload on a new deploy. svelte.config.js polls _app/version.json
   (60 s); the root layout reloads itself when the deployed build changes —
   unless an upload is in flight — so an open tab can't keep running stale code
   after a rebuild (the recurring "my fix isn't applied" trap).

npm run check: 0 errors, 58 tests pass.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

frontend/static/locales/en.json

@@ -503,6 +503,7 @@
     "uploaded": "Upload complete",
     "uploaded_saved": "Upload complete — {{mb}} MB deduplicated",
     "uploaded_partial": "{{ok}} uploaded, {{failed}} failed",
+    "uploaded_skipped": "{{ok}} uploaded · {{skipped}} skipped (not regular files)",
     "upload_failed": "Upload failed",
     "uploading": "Uploading…",
     "uploading_file": "Uploading {{name}}…",

frontend/static/locales/es.json

@@ -369,6 +369,7 @@
     "selected_count": "{{count}} seleccionados",
     "uploaded_saved": "Subida completa — {{mb}} MB deduplicados",
     "uploaded_partial": "{{ok}} subidos, {{failed}} fallaron",
+    "uploaded_skipped": "{{ok}} subidos · {{skipped}} omitidos (no son ficheros normales)",
     "upload_failed": "La subida falló"
   },
   "dialogs": {