We need to decide what to do about these before we release. Supersedes #105.

Here's my current thinking, feel free to modify or suggest complete alternatives:

• Keep dwi2response, but rename it to dwi2sf, and output the single-fibre voxel mask by default; only give the final response function if explicitly requested by the user.
• Rename sh2response to sf2response, and receive as primary input a DWI rather than SH; this will additionally allow for selecting the shell(s) for which response(s) should be estimated (which should simplify msdwi2response).

Hopefully that makes it clear that there are in fact two steps: single-fibre voxel selection, and response function estimation based on those voxels. Just using the (optional) response function output from dwi2sf is a shortcut that will probably satisfy most users, but hopefully the fact that the SF voxel mask is the default output will make people question that and look into alternative ways to get the SF mask rather than just reporting that they can't get a response function.

Hopefully it also separates msdwi2response from what is currently dwi2response, as their operations are actually quite different currently.