fix(HEL-459): 影子比较按请求字段投影，盘后整批原子发布

比较侧只对网站本次请求字段计业务差异，忽略数据中枢额外列；
盘后 A/B/重发改为先整批暂存与交叉校验，再单事务切换公开版本。

Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: multica-agent <github@multica.ai>

backend/data/datahub/compare.py

@@ -5,6 +5,7 @@ from typing import Any
 from backend.data.datahub.native import SCALE_FIELDS, row_key, to_canonical_row, yyyymmdd
 
 NUMERIC_TOLERANCE = 1e-4
+CANONICAL_ALIASES = {"volume": "vol"}
 
 
 def compare_rows(
@@ -13,8 +14,10 @@ def compare_rows(
     hub_rows: list[dict[str, Any]] | None,
     hub_meta: dict[str, Any] | None = None,
     hub_error: str | None = None,
+    fields: str = "",
 ) -> dict[str, Any]:
     hub = hub_rows or []
+    requested = _requested_fields(fields)
     legacy_map = {row_key(dataset, row): row for row in legacy_rows}
     hub_map = {row_key(dataset, _align_hub_row(row)): row for row in hub}
     missing_hub = sorted(key for key in legacy_map if key not in hub_map)
@@ -26,7 +29,7 @@ def compare_rows(
         hub_row = hub_map.get(key)
         if hub_row is None:
             continue
-        field_report = _compare_fields(dataset, legacy, hub_row)
+        field_report = _compare_fields(dataset, legacy, hub_row, requested)
         if field_report["unit_conversion"]:
             unit_conversion.append({"key": list(key), "fields": field_report["unit_conversion"]})
         if field_report["value_diff"]:
@@ -53,6 +56,7 @@ def compare_rows(
         "published_at": (hub_meta or {}).get("published_at"),
         "trade_date": yyyymmdd((hub_meta or {}).get("trade_date")),
         "hub_error": hub_error,
+        "fields_compared": sorted(requested) if requested is not None else None,
         "equal": (
             not hub_error
             and not missing_hub
@@ -71,13 +75,36 @@ def _align_hub_row(row: dict[str, Any]) -> dict[str, Any]:
     return aligned
 
 
-def _compare_fields(dataset: str, legacy: dict[str, Any], hub: dict[str, Any]) -> dict[str, list[dict[str, Any]]]:
+def _requested_fields(fields: str) -> list[str] | None:
+    """Fields the website actually asked for; None means "no projection"."""
+    keys = [item.strip() for item in str(fields or "").split(",") if item.strip()]
+    if not keys:
+        return None
+    seen: list[str] = []
+    for key in keys:
+        canonical = CANONICAL_ALIASES.get(key, key)
+        if canonical not in seen:
+            seen.append(canonical)
+    return seen
+
+
+def _compare_fields(
+    dataset: str,
+    legacy: dict[str, Any],
+    hub: dict[str, Any],
+    requested: list[str] | None = None,
+) -> dict[str, list[dict[str, Any]]]:
     canonical_legacy = to_canonical_row(dataset, legacy)
     hub_canonical = _hub_canonical(dataset, hub)
     native_hub = _align_hub_row(hub)
     value_diff: list[dict[str, Any]] = []
     unit_conversion: list[dict[str, Any]] = []
     keys = (set(canonical_legacy) | set(hub_canonical)) - {"batch_id", "updated_at", "volume"}
+    if requested is not None:
+        # Compare only what the website asked for. Extra hub columns are
+        # transport detail, not business differences; a requested field still
+        # alarms when it is missing or holds a different value.
+        keys = set(requested) - {"batch_id", "updated_at", "volume"}
     scales = SCALE_FIELDS.get(dataset) or {}
     for field in sorted(keys):
         left = canonical_legacy.get(field)