perf(cadastre): freshness skip + grid-walk heartbeat + reduce WMS log noise

Three operational improvements after ekb_full v3 completed at 99.1%:

Fix H — per-quarter freshness skip (scrape_cadastre.py):
- enqueue_cadastre_harvest queries cad_quarter_stats.fetched_at per quarter
- Skip quarters where fetched_at >= NOW() - skip_fresh_hours (default 1h)
- extra_config.skip_fresh_hours=0 disables skip (full re-fetch)
- Saves ~30k NSPD calls on repeated runs of same scope

Fix I — heartbeat in grid-walk loop (bulk_harvest.py:_grid_walk_category):
- 225 cells x 70-300ms = 15-75s heartbeat silence
- Pass update_progress callback to _grid_walk_category
- Call every heartbeat_every (50) cells with phase progress info
- Prevents cleanup_zombies false-positive cancellation

Fix J — NSPD WMS HTTP 500 noise to debug (bulk_harvest.py:_grid_walk_category):
- NSPD WMS endpoint frequently returns HTTP 500 (server-side issue)
- Was: logger.warning each, 200+ warnings per quarter
- Now: logger.debug (visible via --log-level=DEBUG)

40 tests passing.

backend/app/services/cadastre/bulk_harvest.py

@@ -218,6 +218,7 @@ async def harvest_quarter(
             client=client,
             quarter=quarter,
             layer_id=cat_id,
+            update_progress=update_progress,
         )
         result.grid_walk_requests += n_requests
 
@@ -276,12 +277,19 @@ async def _grid_walk_category(
     layer_id: int,
     grid_size: int = 15,
     tile_size: int = 512,
+    update_progress: Callable[[dict[str, Any]], None] | None = None,
+    heartbeat_every: int = 50,
 ) -> tuple[int, int]:
     """Grid-walk layer_id в bbox квартала.
 
     Генерирует grid_size×grid_size ячеек, для каждой делает wms_feature_info.
     Дедуплицирует по cad_num, upsert'ит новые объекты.
 
+    Args:
+        update_progress: Fix I — callback для heartbeat внутри grid-walk loop.
+            225 cells × ~70ms = ~15-75s без heartbeat, zombie-cleanup ложно
+            помечал worker как мёртвого. Вызываем каждые heartbeat_every cells.
+
     Returns:
         (upserted_count, requests_count)
     """
@@ -300,7 +308,19 @@ async def _grid_walk_category(
     upserted = 0
     requests = 0
 
-    for cell_bbox, click_xy in grid_points:
+    for idx, (cell_bbox, click_xy) in enumerate(grid_points):
+        # Fix I: heartbeat каждые N cells — обновляет cadastre_jobs.heartbeat_at
+        # чтобы cleanup_zombies (5min stale check) не cancel'нул живой worker.
+        if update_progress is not None and idx > 0 and idx % heartbeat_every == 0:
+            update_progress(
+                {
+                    "phase": f"grid_walk_{layer_id}_cell_{idx}",
+                    "quarter": quarter,
+                    "grid_cells_done": idx,
+                    "grid_cells_total": len(grid_points),
+                }
+            )
+
         try:
             features = await client.wms_feature_info(
                 layer_id=layer_id,
@@ -311,7 +331,10 @@ async def _grid_walk_category(
             )
             requests += 1
         except Exception as e:
-            logger.warning(
+            # Fix J: NSPD WMS возвращает HTTP 500 на десятки cells per quarter —
+            # это server-side noise, не наш bug. logger.debug чтобы не засорять
+            # prod логи (раньше было warning → 200+ warnings per quarter).
+            logger.debug(
                 "_grid_walk_category: wms_feature_info error layer=%d quarter=%s: %s",
                 layer_id,
                 quarter,

backend/app/workers/tasks/scrape_cadastre.py

@@ -357,22 +357,65 @@ def enqueue_cadastre_harvest(self: Any, job_id: int) -> dict[str, Any]:
             logger.warning("enqueue_cadastre_harvest: job=%d имеет пустой список quarters", job_id)
             return {"job_id": job_id, "enqueued": 0}
 
+        # Fix H: per-quarter freshness skip — не переобрабатывать quarters с свежим
+        # cad_quarter_stats.fetched_at (последний успешный обход <FRESHNESS_HOURS).
+        # extra_config.skip_fresh_hours=N управляет порогом; 0 → отключает skip.
+        # Это сэкономит ~30k NSPD calls при повторных запусках того же scope.
+        extra_config = job["extra_config"] or {}
+        skip_fresh_hours = extra_config.get("skip_fresh_hours", 1)
+        skipped_fresh: set[str] = set()
+        if skip_fresh_hours > 0:
+            threshold = _dt.datetime.now(_dt.UTC) - _dt.timedelta(hours=skip_fresh_hours)
+            rows = (
+                db.execute(
+                    text(
+                        "SELECT cad_number FROM cad_quarter_stats "
+                        "WHERE cad_number = ANY(CAST(:qs AS text[])) "
+                        "AND fetched_at >= :t"
+                    ),
+                    {"qs": quarters, "t": threshold},
+                )
+                .scalars()
+                .all()
+            )
+            skipped_fresh = set(rows)
+            if skipped_fresh:
+                # Корректируем targets_total — skipped quarters не входят в прогресс
+                db.execute(
+                    text(
+                        "UPDATE cadastre_jobs SET targets_total = :new_total, "
+                        "targets_skipped = COALESCE(targets_skipped, 0) + :sk "
+                        "WHERE job_id = :id"
+                    ),
+                    {
+                        "new_total": len(quarters) - len(skipped_fresh),
+                        "sk": len(skipped_fresh),
+                        "id": job_id,
+                    },
+                )
+
         # Обновить статус на running
         _update_job(db, job_id, status="running", started_at=_dt.datetime.now(_dt.UTC))
         db.commit()
 
+        enqueued = 0
         for quarter in quarters:
+            if quarter in skipped_fresh:
+                continue
             bulk_harvest_quarter_task.apply_async(
                 args=[quarter, job_id],
                 queue=CADASTRE_QUEUE,
             )
+            enqueued += 1
 
         logger.info(
-            "enqueue_cadastre_harvest: job=%d enqueued %d quarters",
+            "enqueue_cadastre_harvest: job=%d enqueued %d (skipped_fresh=%d, threshold=%dh)",
             job_id,
-            len(quarters),
+            enqueued,
+            len(skipped_fresh),
+            skip_fresh_hours,
         )
-        return {"job_id": job_id, "enqueued": len(quarters)}
+        return {"job_id": job_id, "enqueued": enqueued, "skipped_fresh": len(skipped_fresh)}
 
     finally:
         db.close()