Record implicitly converted dataset as input dataset

I believe this was always the intention and is how it works if the
dataset already exists (i.e. on a re-run). The converter records the
dataset that the user chose, so there's no gap in the provenance either.

This somewhat addresses
galaxyproject/total-perspective-vortex#141
so you can (reliably) differetiate your rules on the input datatype
and filesize combination.

lib/galaxy/tools/actions/__init__.py

@@ -139,7 +139,9 @@ def process_dataset(data, formats=None):
                     if converted_dataset:
                         data = converted_dataset
                     else:
-                        data = data.get_converted_dataset(trans, target_ext, target_context=parent, history=history)
+                        data = data.get_converted_dataset(
+                            trans, target_ext, target_context=parent, history=history, return_output=True
+                        )
 
                 input_name = prefixed_name
                 # Checked security of whole collection all at once if mapping over this input, else