ENH: Separate Space and SpatialReferences

[oesteban]

Okay, after much thinking, I believe this captures the whole set of requirements.

Now there is an Space object that makes all the validations and sanity checks. There is still room for enhancements:

• We might want to implement a classmethod (something like Space.from_string) that implements the parsing of --output-spaces arguments (see https://www.attrs.org/en/stable/init.html). DONE
• We might want to check that the specs contain valid keywords against templateflow (at the moment, this is only done for cohort).

Other than that, I think the doctest examples show pretty much all the functionality.

Then, the SpatialReferences class is the one in charge of sorting out the available spaces. It has three convenience methods to filter out references:

• get_std_spaces() returns template names (including cohort if necessary) that is of interest to the spatial normalizations that sMRIPrep will run.
• get_templates() returns full Space objects with the desired outputs, as set by the user.
• get_nonstd_spaces() returns nonstandard Spaces or their names (depending on the only_names argument)

[oesteban]

Added tests on examples like those in: nipreps/fmriprep#1604 (comment)

[oesteban]

Moving to niworkflows, since the changes are quite massive - I guess there's no point in looking at the diff.