Mask preprocessing in registration

[cookpa]

While reviewing #450, I saw this

ANTsPy/ants/registration/interface.py

Lines 396 to 400 in 8f5fa94

	if mask is not None:
	mask_scale = mask - mask.min()
	mask_scale = mask_scale / mask_scale.max() * 255.0
	charmask = mask_scale.clone("unsigned char")
	maskopt = "[%s,NA]" % (utils.get_pointer_string(charmask))

This might make masks be interpreted differently to ANTs, if they are not binarized already. For example, if one used a FreeSurfer aparc+aseg type mask here ( subcortex labels 1-48(ish), cortex 1000-1100(ish) ), some labels will be zeroed out in this rescaling process.

How about we just binarize first and then cast to char? Happy to work on a PR for this.

[ntustison]

Thanks Phil. Yeah, in all my coding, I go with ITK's definition of a mask which is anything not equal to 0.