ENH: support resampling multi-channel images

[ncullen93]

This PR adds support for multi-channel images in resample_image. The docstring says it's supported and there has pretty much always been an exception raised for multi-channel, so I think I just never got around to implementing it.

This implementation is the simplest - it just splits the channels and resamples on each of them individually. I think the ants function ResampleImage supports vector images natively, so that may be preferred in the future.

[coveralls]

coverage: 55.357% (+0.1%) from 55.245% when pulling 6f4fa76f2d5d9168106b234ad7a25d00532f487c on resample-channels into 54ec16ef994273b409207628b56ba7ab3c825383 on master.

[cookpa]

ANTs ResampleImage is templated over dimension so it will work for 3D images or 4D time series. Your approach of resampling each channel separately makes more sense for actual multi-channel images, and things like DWI data that are sometimes stored on disk as time series, but where it does not really make sense to interpolate along the fourth dimension.

I don't know how best to harmonize these things, in antsApplyTransforms we have different codes for 4D (time series) or 5D (vectors / tensors / generic multi-channel with 3 spatial dimensions and single time value).

[cookpa]

Looking at the function I think it might actually be good, becuase antspy will recognize components in a 5D image, and if components is >1 then it calls the new code, and if components is 1 it will call ResampleImage which already works for time series.

[ncullen93]

OK - thanks for the input. I will merge but feel free to open an issue in the future if needed and I will change it.