AlignmentDUTResidual has erratic behaviour with GBL tracking

When we use General Broken Lines for our tracking, after aligning the telescope with AlignmentTrackChi2 and doing a rough prealignment using Correlations, running AlignmentDUTResidual produces some very wild deltas in the DUT rotation angles and does not converge (even though we have a pretty high efficiency over the sensor before doing it). I spoke with @lhuth and the issue seems to go away when we revert to straight line tracking.

If it isn't reproducible with other data I can provide mine to help reproduce.

cc @lhuth