FR: Arbitrary output resolution (for --output-spaces)

[oesteban]

Follows up on #1604. After merging #1955, there is still one use-case we haven't covered: allowing arbitrary resolutions.

E.g., MNI152Lin:res-iso1.6mm

This issue reminds us of:

1. Discuss whether this feature should be added
2. If so, then finalize its implementation.

[effigies]

This issue came up for anatomical (not template) here: https://neurostars.org/t/fmriprep-different-resolution-between-bold-files-in-space-t1w-and-actual-t1w/19231/2

I think there is a valid argument to be made for upsampling, as it preserves more of the signal that's being resampled. e.g., consider the worst possible case of shifting everything by 0.5 voxel in every direction, and each resulting voxel is simply an average of the 8 voxels surrounding the corner it's being mapped onto.