Use explicit reason for dataset type subsetting failing

Report the missing dataset type rather than a more
opaque default KeyError.

Also allow the incompatibility to be accepted if the storage
classes are compatible.

This can happen if a dataset type definition defined in a
connections class that is not yet in registry, does not match the
definition added elsewhere. The example that triggered this
was TaskMetadata which was used in a connection and added
by the pipeline infrastructure.

python/lsst/pipe/base/graphBuilder.py

@@ -125,9 +125,44 @@ def fromSubset(
             A new dictionary instance.
         """
         combined = ChainMap(first, *rest)
-        return cls(
-            {datasetType: combined[datasetType] for datasetType in datasetTypes}, universe=first.universe
-        )
+        _dict = {}
+        for datasetType in datasetTypes:
+            if datasetType in combined:
+                _dict[datasetType] = combined[datasetType]
+            else:
+                # The dataset type is not found. It may not be listed
+                # or it may be that it is there with the same name
+                # but different definition.
+                for d in combined:
+                    if d.name == datasetType.name:
+                        # This implies some inconsistency in definitions
+                        # for connections. If there is support for storage
+                        # class conversion we can let it slide.
+                        # At this point we do not know
+                        # where the inconsistency is but trust that down
+                        # stream code will be more explicit about input
+                        # vs output incompatibilities.
+                        if d.is_compatible_with(datasetType) or datasetType.is_compatible_with(d):
+                            _LOG.debug(
+                                "Dataset type mismatch (%s != %s) but continuing since they are compatible",
+                                datasetType,
+                                d,
+                            )
+                            _dict[datasetType] = combined[d]
+                            break
+                else:
+                    raise KeyError(
+                        "Dataset type definition mismatch and they are not compatible: "
+                        f"{datasetType} != {d}"
+                    )
+
+                if datasetType not in _dict:
+                    raise KeyError(
+                        f"DatasetType {datasetType} not present in list of known types: "
+                        + ", ".join(d.name for d in combined)
+                    )
+
+        return cls(_dict, universe=first.universe)
 
     @property
     def dimensions(self) -> DimensionGraph: