feat(suggestions): tag-input dropdown searches the full prediction set

The typed dropdown sourced the threshold-filtered panel list (>= 0.70 general),
so low-confidence actions/features the model DID predict never appeared — forcing
hand-typed custom tags instead of accepting the model's canonical formatting.

Add a threshold override: SuggestionService.for_image(threshold_override=) and
GET /images/<id>/suggestions?min=<f> surface EVERY stored prediction (down to the
0.05 store floor), alias-resolved and normalized, still excluding applied/rejected
and unsurfaced categories. The suggestions store gains allByCategory + loadAll
(min=0); the dropdown searches that full set (cap 20), while the Suggestions panel
stays curated at the configured threshold. Accept/dismiss drop from both lists.

Operator-asked 2026-06-09. Test: a 0.30 general prediction is hidden by default
but surfaced with threshold_override=0.0; unsurfaced categories still excluded.

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

frontend/src/components/modal/SuggestionsPanel.vue

@@ -62,7 +62,11 @@ const isEmpty = computed(() =>
   Object.values(store.byCategory).every(list => !list || list.length === 0)
 )
 
-watch(() => props.imageId, (id) => { if (id != null) store.load(id) }, { immediate: true })
+watch(() => props.imageId, (id) => {
+  if (id == null) return
+  store.load(id)        // panel: curated, ≥ threshold
+  store.loadAll(id)     // dropdown: full prediction set (low-confidence included)
+}, { immediate: true })
 
 // After a successful accept/alias-accept, refresh the modal's current
 // tag list so TagPanel's chip rail reflects the newly-attached tag.

frontend/src/components/modal/TagAutocomplete.vue

@@ -190,12 +190,17 @@ function scorePct (s) { return `${Math.round(s.score * 100)}%` }
 
 // This image's suggestions that match the typed query, minus any the server
 // autocomplete already returned (same name+kind) so a tag never shows twice.
+// Sources the FULL prediction set (allByCategory, down to the store floor) — NOT
+// the threshold-filtered panel list — so a low-confidence action/feature the
+// model saw can be typed and accepted in canonical formatting instead of being
+// hand-entered as a custom tag (operator-asked 2026-06-09). The typed query is
+// the only filter; the threshold no longer hides anything here.
 const suggestionHits = computed(() => {
   const q = parsedName.value.toLowerCase()
   if (!q) return []
   const seen = new Set(hits.value.map(h => `${h.kind}:${h.name.toLowerCase()}`))
   const out = []
-  for (const list of Object.values(suggestions.byCategory)) {
+  for (const list of Object.values(suggestions.allByCategory)) {
     for (const s of list || []) {
       const key = `${s.category}:${s.display_name.toLowerCase()}`
       if (!s.display_name.toLowerCase().includes(q)) continue
@@ -204,9 +209,10 @@ const suggestionHits = computed(() => {
       out.push(s)
     }
   }
-  // Best matches first; cap so the dropdown stays scannable.
+  // Best matches first; cap generously so a specific typed query surfaces its
+  // matches even when many predictions exist, while the list stays scrollable.
   out.sort((a, b) => b.score - a.score)
-  return out.slice(0, 6)
+  return out.slice(0, 20)
 })
 
 // One ordered list backing both the rendered dropdown and keyboard nav, so the