feat(settings): tag-eval admin card — trigger + persisted report (survives nav)

Frontend for #1130. A maintenance tile in Settings → Tagging:
- Editable concept list + "Run eval" → POST /api/tag-eval (one running at a time).
- Rehydrates on mount via the persisted run (getRun by latest id) and polls while
  running — so the report SURVIVES navigation (operator-flagged); the task runs
  backend-side regardless and the card reconnects to its row.
- Renders the saved report: per-concept head-vs-centroid metrics table (AP/F1/
  precision/recall) with Δ AP, the learning curve (AP @ N positives), and
  thumbnail galleries (head-would-suggest / head-doubts-positive) for eyeballing.

Backend: _examples now stores thumbnail_urls (not just ids) so the report is a
self-contained artifact that renders without per-id lookups on reload.

No new top-level surface — slots into the existing maintenance area.

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

backend/app/services/ml/tag_eval.py

@@ -201,7 +201,7 @@ def _eval_concept(session: Session, name: str, cfg: dict, np) -> dict[str, Any]:
     head = _eval_head(Xn, y, cfg["cv_folds"], np)
     centroid = _eval_centroid(Xn, y, cfg["cv_folds"], np)
     curve = _learning_curve(Xn, y, cfg["curve_points"], neg_ratio, np)
-    examples = _examples(Xn, y, ids, np)
+    examples = _examples(session, Xn, y, ids, np)
 
     return {
         "name": name, "tag_id": tag_id,
@@ -296,11 +296,13 @@ def _learning_curve(Xn, y, points, neg_ratio, np) -> list[dict[str, float]]:
     return out
 
 
-def _examples(Xn, y, ids, np) -> dict[str, list[int]]:
+def _examples(session, Xn, y, ids, np) -> dict[str, list[dict]]:
     """Train on all data, then surface: top-scoring UNLABELED-ish (highest among
     the negative pool = what the head would newly suggest) and lowest-scoring
     POSITIVES (where the head disagrees with the operator's tag — likely the
-    most informative to review)."""
+    most informative to review). Resolves thumbnail urls so the stored report
+    renders without per-id lookups (survives navigation as a self-contained
+    artifact)."""
     from sklearn.linear_model import LogisticRegression
 
     clf = LogisticRegression(max_iter=1000, class_weight="balanced")
@@ -308,9 +310,26 @@ def _examples(Xn, y, ids, np) -> dict[str, list[int]]:
     s = clf.predict_proba(Xn)[:, 1]
     neg_idx = np.where(y == 0)[0]
     pos_idx = np.where(y == 1)[0]
-    top_neg = neg_idx[np.argsort(s[neg_idx])[::-1][:_EXAMPLES_K]]
-    low_pos = pos_idx[np.argsort(s[pos_idx])[:_EXAMPLES_K]]
+    top_neg = [int(ids[i]) for i in neg_idx[np.argsort(s[neg_idx])[::-1][:_EXAMPLES_K]]]
+    low_pos = [int(ids[i]) for i in pos_idx[np.argsort(s[pos_idx])[:_EXAMPLES_K]]]
+    thumbs = _resolve_thumbs(session, top_neg + low_pos)
     return {
-        "head_would_suggest": [int(ids[i]) for i in top_neg],
-        "head_doubts_positive": [int(ids[i]) for i in low_pos],
+        "head_would_suggest": [thumbs[i] for i in top_neg if i in thumbs],
+        "head_doubts_positive": [thumbs[i] for i in low_pos if i in thumbs],
     }
+
+
+def _resolve_thumbs(session, ids: list[int]) -> dict[int, dict]:
+    from ..gallery_service import thumbnail_url
+
+    out: dict[int, dict] = {}
+    if not ids:
+        return out
+    for rid, tp, sha, mime in session.execute(
+        select(
+            ImageRecord.id, ImageRecord.thumbnail_path,
+            ImageRecord.sha256, ImageRecord.mime,
+        ).where(ImageRecord.id.in_(ids))
+    ).all():
+        out[rid] = {"id": rid, "thumbnail_url": thumbnail_url(tp, sha, mime)}
+    return out