multi_model_statistics fail when only a single dataset is given

[schlunma]

(Apologies for posting this again in case this was already discussed, but I didn't find an issue about this.)

While testing #673 I realized that multi_model_statistics fails when only a single input dataset is given:

ESMValCore/esmvalcore/preprocessor/_multimodel.py

Lines 294 to 296 in 9c008ee

	if len(cubes) == 1:
	raise ValueError('Cannot perform multicube statistics '
	'for a single cube.')

Especially with the introduction of groupy and ensemble_statistics, I don't think this is good behavior. For most statistics we support there are reasonable default values (i.e. the formulas used to calculate these values work perfectly fine for just one value), e.g. simply return the input data for mean, gmean, hmean, min, max, sum, rms; or return 0 for std, var. I'm not entirely sure about others like percentiles.

Would it make sense to just "hardcode" these defaults for these special cases? Unfortunately iris also raises an error in this case:
iris.exceptions.CoordinateCollapseError: Cannot collapse a dimension which does not describe any data. @ESMValGroup/esmvaltool-coreteam

Note: It is possible to avoid this by using the exclude key, but since this only takes dataset names into account it might fail for more complex ensembles and groupby settings.

[schlunma]

Duplicate of #1211