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/bridge.py

@@ -122,10 +122,12 @@ class DatahubBridge:
                 legacy_rows = legacy_query(api_name, params, fields)
             except Exception as exc:
                 if flags.read and hub_rows is not None and hub_error is None:
-                    self._emit_shadow(compare_rows(dataset, [], hub_canonical, hub_meta, self._error_text(exc)))
+                    self._emit_shadow(
+                        compare_rows(dataset, [], hub_canonical, hub_meta, self._error_text(exc), fields)
+                    )
                     return project_fields(hub_rows, fields)
                 raise
-            self._emit_shadow(compare_rows(dataset, legacy_rows, hub_canonical, hub_meta, hub_error))
+            self._emit_shadow(compare_rows(dataset, legacy_rows, hub_canonical, hub_meta, hub_error, fields))
             if flags.read and hub_rows is not None and hub_error is None:
                 return project_fields(hub_rows, fields)
             return legacy_rows