Improve handling of Timespan in queries.

Postgres sometimes needs help in guessing column type when doing UNION
of several queries.

python/lsst/daf/butler/registry/databases/postgresql.py

@@ -504,7 +504,11 @@ def extract(cls, mapping: Mapping[str, Any], name: str | None = None) -> Timespa
     def fromLiteral(cls, timespan: Timespan | None) -> _RangeTimespanRepresentation:
         # Docstring inherited.
         if timespan is None:
-            return cls(column=sqlalchemy.sql.null(), name=cls.NAME)
+            # Cast NULL to an expected type, helps Postgres to figure out
+            # column type when doing UNION.
+            return cls(
+                column=sqlalchemy.func.cast(sqlalchemy.sql.null(), type_=_RangeTimespanType), name=cls.NAME
+            )
         return cls(
             column=sqlalchemy.sql.cast(
                 sqlalchemy.sql.literal(timespan, type_=_RangeTimespanType), type_=_RangeTimespanType

python/lsst/daf/butler/registry/datasets/byDimensions/_storage.py

@@ -612,7 +612,7 @@ def make_query_joiner(self, collections: Sequence[CollectionRecord], fields: Set
             )
             if "timespan" in fields:
                 tags_builder.joiner.timespans[self.datasetType.name] = (
-                    self._db.getTimespanRepresentation().fromLiteral(Timespan(None, None))
+                    self._db.getTimespanRepresentation().fromLiteral(None)
                 )
         calibs_builder: QueryBuilder | None = None
         if CollectionType.CALIBRATION in collection_types: