amp2response speedup

[Lestropie]

Pre-allocate memory for ZSH transformations and image data, rather than using Eigen's conservativeResize() function. According to a user, their dwi2response call went from 40 minutes to 8.

Also should fix the segfault on providing an XxYxZx1 mask image as reported in #1231.

Closes #1231.

[jdtournier]

According to a user, their dwi2response call went from 40 minutes to 8.

😮

Guess we need to fix the testing first though... I'll work on that today.

[jdtournier]

Let's see if the recent merge of #1393 will help with the testing here...

In the meantime, any chance you could take me through the maths behind amp2response? Maybe they're already described in some other issue? I'm just surprised it would still take 8 minutes - I reckon there's scope for some further optimisation, but I'd like to get a better feel for what you're trying to do before suggesting something daft...

[Lestropie]

That timing would have been for dwi2response as a whole, not just amp2response.

Re-acquired some HCP data to check it myself: dwi2response dhollander was 2:40 on dev, 2:10 on this branch. Guessing that if the system starts swapping at all the difference will be more drastic.

Also confirm that this worked with a XxYxZx1 mask image, whereas dev segfaulted; should push that to master as well I think.