M1-rework: rename leftover pk_cols param in reconciliation test stubs

Round-2 audit nit (review-notes/M1-full-review-2.md): two _always_fail
monkeypatch stubs still named their (ignored) third positional parameter
pk_cols after _reconcile switched to a full columns list. Rename to columns
for consistency. Test-only, no behavior change.

pytest 97 passed; ruff clean.

tests/test_migrate_legacy_data.py

@@ -277,7 +277,7 @@ def test_migrate_reconciliation_failure_raises(tmp_path: Path, monkeypatch: pyte
     _make_legacy_location_db(legacy_path, LOCATION_ROWS)
     legacy_url = f"sqlite:///{legacy_path}"
 
-    def _always_fail(conn, table, pk_cols, source_count):
+    def _always_fail(conn, table, columns, source_count):
         # Simulate a scenario where reconciliation finds rows missing
         raise RuntimeError(
             f"Reconciliation failed for table '{table}': "
@@ -302,7 +302,7 @@ def test_cli_exits_nonzero_on_reconciliation_failure(
     legacy_url = f"sqlite:///{legacy_path}"
 
     # Patch _reconcile to always raise
-    def _always_fail(conn, table, pk_cols, source_count):
+    def _always_fail(conn, table, columns, source_count):
         raise RuntimeError(
             f"Reconciliation failed for table '{table}': 1 row missing."
         )