amp2response performance

[Lestropie]

As raised here (even though there's an upstream problem in that particular case).

• Don't bother constructing transformation matrices when using the -isotropic option.

• Pre-determine the number of voxels in the mask, and pre-allocate cat_transforms and cat_data rather than using conservativeResize().

[thijsdhollander]

While this sits here, just adding something I forgot to create an issue for, but stumbled upon at some point: if I remember well, amp2response crashes (segfault) when the provided mask (i.e. second argument) is 4-dimensional, even with the size of the 4th dimension being "1" (so the mask essentially being 3D).