Refine options for handling of mismatches in metric dimensions

[jbusecke]

The handling of errors during the "metric-matching" step is still not great:

1. The docstrings of postprocessing.match_metrics and postprocessing._parse_metric need to include all valid options
2. I would like another option 'ignore', which simply skips the dataset in question (this might have to be implemented one level higher in postprocessing.match_metrics. Perhaps something like include_no_match which if False, skips both datasets with no match and errors (this might be neat because the errors could still be printed?).

Prominent example that always seems to trigger this is one/some of the GISS models which have areas that do not have the same shape as the data.