ch_root: Don't require both the input and output mappings to contain their respective new roots, since sometimes only one or the other root is being subset. This will occur, for example, in mappings that are flat on the input but normalized on the output, such as VegCSV.

VegBIEN: Reversed aggregateoccurrence<->plantobservation relationship to point from plantobservation->aggregateoccurrence, so plantobservation could be scoped by aggregateoccurrence in the same way as all other core tables are scoped by their parent tables. This reversed direction was an anomaly due to the need to have a trigger auto-set aggregateoccurrence.count to 1 when there was an associated plantobservation. This was most easily accomplished on the aggregateoccurrence table itself, but required the reversed relationship. The trigger has now been reimplemented on plantobservation, which externally updates aggregateoccurrence.count.

input.Makefile: Testing: diffing test outputs: Ignore changes in whitespace, due to e.g. different indent levels. This facilitates accepting tests when an element has been nested inside another element (or unnested), by showing only the opening and closing tags of the new outer element.

dicts.py: DictProxy: Fixed bug where default value for inner param needed to be created in the constructor, or else every default instance would use and modify the same dictionary

db_xml.py: put(): wrap_e(): Call augment_error() to add the current node to the error message

db_xml.py: put(): Raise an error if there are multiple fields with the same name, instead of silently overwriting the first with the second. This generally indicates the need to use `:[@merge=1]` on the fields in question.

dicts.py: Added OnceOnlyDict and helper exception KeyExistsError

dicts.py: DictProxy: Added default value for inner param to facilitate creating empty wrapped dicts

bin/map: out_is_db: row-based mode: Debug-log the processed XML tree produced by xml_func.process()

sql_io.py: put_table(): Fixed bug where Missing mapping for NOT NULL column errors should actually be warnings because sometimes the mappings include extra tables which aren't used by the dataset