feat: online verification, node removals, and knowledge browser

Adds claim-bound online evidence records and citation verification scripts,
node removal transport/adapter, retrieval benchmark and owner-review tooling,
GraphCanvas/SourceDetail knowledge browser UI, and the ingest handoff helper.
Bumps the agents/gewu pack to the online-verification branch.

tests/test_score_retrieval_benchmark.py

@@ -0,0 +1,74 @@
+import json
+from pathlib import Path
+
+import pytest
+from scripts.score_retrieval_benchmark import score
+
+
+def _inputs(tmp_path: Path) -> tuple[Path, Path]:
+    results = tmp_path / "results.jsonl"
+    rows = [
+        {
+            "kind": "manifest",
+            "question_count": 2,
+            "questions_sha256": "frozen-hash",
+        },
+        {
+            "kind": "retrieval",
+            "question": {"id": "Q01"},
+            "seconds": 0.2,
+            "is_error": False,
+            "returned_node_ids": ["node-a", "other"],
+            "response": [{"canonical_revision": "a" * 40}],
+        },
+        {
+            "kind": "retrieval",
+            "question": {"id": "Q02"},
+            "seconds": 0.4,
+            "is_error": False,
+            "returned_node_ids": ["node-c"],
+            "response": [{"canonical_revision": "a" * 40}],
+        },
+    ]
+    results.write_text("".join(json.dumps(row) + "\n" for row in rows), encoding="utf-8")
+    judgments = tmp_path / "judgments.json"
+    judgments.write_text(
+        json.dumps(
+            {
+                "method": "independent relevance review",
+                "fully_answerable_positive_questions": {
+                    "Q01": ["node-a", "node-b"],
+                    "Q02": ["node-c"],
+                },
+            }
+        ),
+        encoding="utf-8",
+    )
+    return results, judgments
+
+
+def test_scores_recall_on_judged_answerable_questions_only(tmp_path: Path) -> None:
+    results, judgments = _inputs(tmp_path)
+
+    result = score(results, judgments)
+
+    assert result["question_count"] == 2
+    assert result["judged_answerable_questions"] == 2
+    assert result["mean_conditional_recall_at_5"] == 0.75
+    assert result["conditional_hit_rate"] == 1.0
+    assert result["canonical_revision"] == "a" * 40
+
+
+def test_rejects_missing_judged_questions_and_mixed_canonical_heads(tmp_path: Path) -> None:
+    results, judgments = _inputs(tmp_path)
+    rows = results.read_text(encoding="utf-8").splitlines()
+    missing_results = tmp_path / "missing.jsonl"
+    missing_results.write_text("\n".join(rows[:2]) + "\n", encoding="utf-8")
+    with pytest.raises(ValueError, match="count does not match"):
+        score(missing_results, judgments)
+
+    rows[2] = rows[2].replace("a" * 40, "b" * 40)
+    mixed = tmp_path / "mixed.jsonl"
+    mixed.write_text("\n".join(rows) + "\n", encoding="utf-8")
+    with pytest.raises(ValueError, match="more than one canonical revision"):
+        score(mixed, judgments)