Fix compare load crash from ORM yield_per with unique().

Load metric rows in LIMIT/OFFSET chunks so large-table compares no longer hit the SQLAlchemy yield_per/unique conflict.

Co-authored-by: Cursor <cursoragent@cursor.com>

netx_api/biz_state/compare_service.py

@@ -1348,15 +1348,22 @@ def _port_map_dict(db: Session, mapping_id: str) -> dict[str, str]:
 
 
 def _load_metric_rows(db: Session, *, batch_id: str, metric_id: str) -> list[dict[str, Any]]:
-    """Load metric rows with streaming fetch to avoid holding the full ORM set."""
+    """Load metric rows in LIMIT/OFFSET chunks (avoid ORM yield_per + unique() clash)."""
     if metric_id == "lldp_neighbor":
-        q = (
+        out: list[dict[str, Any]] = []
+        offset = 0
+        while True:
+            chunk = (
                 db.query(BizStateLldpNeighbor)
                 .filter(BizStateLldpNeighbor.batch_id == batch_id)
-            .execution_options(stream_results=True, yield_per=_LOAD_YIELD_PER)
+                .order_by(BizStateLldpNeighbor.id.asc())
+                .offset(offset)
+                .limit(_LOAD_YIELD_PER)
+                .all()
             )
-        out: list[dict[str, Any]] = []
-        for n in q:
+            if not chunk:
+                break
+            for n in chunk:
                 out.append(
                     {
                         "local_if": n.local_if,
@@ -1377,21 +1384,30 @@ def _load_metric_rows(db: Session, *, batch_id: str, metric_id: str) -> list[dic
                     }
                 )
                 db.expunge(n)
+            offset += len(chunk)
+            if len(chunk) < _LOAD_YIELD_PER:
+                break
         return out
     # Generic tabular metrics (ISIS / interface / ARP / ND6 / BGP …)
     from ..models import BizStateMetricRow
 
-    q = (
+    out = []
+    offset = 0
+    while True:
+        chunk = (
             db.query(BizStateMetricRow)
             .filter(
                 BizStateMetricRow.batch_id == batch_id,
                 BizStateMetricRow.metric_id == metric_id,
             )
             .order_by(BizStateMetricRow.seq.asc(), BizStateMetricRow.id.asc())
-        .execution_options(stream_results=True, yield_per=_LOAD_YIELD_PER)
+            .offset(offset)
+            .limit(_LOAD_YIELD_PER)
+            .all()
         )
-    out = []
-    for r in q:
+        if not chunk:
+            break
+        for r in chunk:
             # Raw rows only — filtering belongs to the compare sheet template
             # (``row_filters``), not metric-specific branches here.
             # ``_netx`` is collector provenance (stripped before field compare).
@@ -1411,6 +1427,9 @@ def _load_metric_rows(db: Session, *, batch_id: str, metric_id: str) -> list[dic
                 }
             )
             db.expunge(r)
+        offset += len(chunk)
+        if len(chunk) < _LOAD_YIELD_PER:
+            break
     if out:
         return out
     # Known metric with zero rows is OK; unknown metric still errors