[ATLASRECTS-7618] handle case of parallel surface/trajectory in RungeKuttaUtils transformGlobalToLine() to avoid FPEs

It's the same underlying issue as for ATLASRECTS-6841, but for the inverse function, and this has been causing FPEs in a few events per run over the year. I'm proposing a different fix than in !50644 (merged), since in the global -> local direction one can obviously still calculate the distance of closest approach even in the collinear case; although this may not be of any use in practice, as transformGlobalToLine() won't able to recover the correct position from that since the azimuth information is lost.

Let's see if @goetz and @christos find this reasonable.

It's probably FT0V (unless those tracks are discarded in the reconstruction: in the event I tested, the trajectory was along the z axis), but I doubt any test will spot it since the FPEs are pretty rare overall.