First proposed change to default tracking parameters aspart of #1729.

First-order algorithms use a very small default step size in order to mitigate curvature overshoot. An alternative solution to curvature overshoot is to use the 4th-order Runge-Kutta method. Therefore, when the latter is activated using the -rk4 option in tckgen, there is no longer a good justification to keep the default step size very small. However in the current code, the same default step size of 1/10th of a voxel is used regardless of the use of RK4.

Initially I was going to make the default step size with RK4 1/2 of a voxel, mimicking the default for iFOD2. However I then considered that for default iFOD2 parameters, there are in fact 3 vertices per 1/2-voxel "step" that are used for sampling 5TT image / ROIs more densely. If the default step size for these first-order algorithms with RK4 were to be 1/2 the size of a diffusion image voxel, there may be a risk of "jumping" features that are defined at a smaller spatial scale. I therefore went for a compromise of 1/4 of a voxel size.

This brings run times with RK4 closer to that without. The fraction of streamlines retained vs. rejected is generally comparable between no RK4 / RK4 with 1/10th voxel step / RK4 with 1/4 voxel step. As far as results are concerned it's difficult to judge the tractograms as being any different.

(iFOD1 and sd_stream outputs are looking particularly woeful however; the latter is probably related to #1630, may need to look into the former).

If this change is accepted it may alter the considerations within #1630.

There is also a further question in here with respect to the interaction between RK4 and the maximum angle. Internally RK4 involves taking two fibre orientation estimates at half a step size away from the current vertex. Currently, no maximum angle constraint is applied at those two points (with the exception of ifod1, which needs its rejection sampling constrained); the maximum angle constraint is only applied based on the fibre orientation estimate at the new vertex. Question is: Should a restriction of half / all of the maximum angle constraint be applied to those two interior calculation points?