gwshop: count dump coverage by unique objectID (review fixes)

Summed nbHits across facet branches double-counted products listed in
several branches; coverage is now per unique objectID, with a warning
about records left out of the dump. AGENTS.md documents the live
count (4183 records, 3491 via facet).

dlphelper/gwshop.py

@@ -149,7 +149,15 @@ class GwShopClient:
         if total <= hits_per_page:
             return hits
         log.info("gwshop %s: %d hits, partitioning on %s", index, total, self._FACET)
-        return self._partition(index, [], total)
+        out = self._partition(index, [], total)
+        covered = len({h.get("objectID") for h in out})
+        if covered < total:
+            log.warning(
+                "gwshop %s: %d of %d index records left out of the dump "
+                "(no fetchable %s branch path)",
+                index, total - covered, total, self._FACET,
+            )
+        return out
 
     def _partition(
         self, index: str, path: list[tuple[int, str]], total: int
@@ -158,26 +166,28 @@ class GwShopClient:
         filters = self._filter_expr(path) if path else None
         counts = self._facet_counts(index, level, filters)
         out: list[dict] = []
-        matched = 0
+        covered: set[Any] = set()
         for value, count in sorted(counts.items()):
             sub_path = path + [(level, value)]
             sub_filters = self._filter_expr(sub_path)
             if count <= _PAGE_SIZE:
                 hits, nb = self._search_all(index, sub_filters)
                 out.extend(hits)
-                matched += nb
+                covered.update(h.get("objectID") for h in hits)
                 log.info(
                     "gwshop %s: %s -> %d hits", index, sub_filters[:80], nb
                 )
             else:
                 sub = self._partition(index, sub_path, count)
                 out.extend(sub)
-                matched += len(sub)
-        residual = total - matched
+                covered.update(h.get("objectID") for h in sub)
+        # nbHits summed across branches double-counts products listed in
+        # several branches, so coverage is counted per unique objectID.
+        residual = total - len(covered)
         if residual > 0:
             log.warning(
-                "gwshop %s: %d hits do not carry %s.lvl%d "
-                "(unclassified products, left out of the dump)",
+                "gwshop %s: %d hits in branch carry no %s.lvl%d "
+                "(shallow facet path, left out of the dump)",
                 index, residual, self._FACET, level,
             )
         return out