Moving Least Squares Curve Fitting Results are different depending on the build environment?

[tavaughan]

Consider this point set:
Points.zip

Results of using Moving Least Squares curve fitting on my local build of Slicer (a bit old, from early July):

Results of using Moving Least Squares curve fitting on nightly Slicer build from Sept 4:

Note that this is very similar to the Global Least Squares curve fit (ends are slightly different).

[tavaughan]

Important part of solution is to add automated tests that run nightly, may help catch similar bugs and pinpoint the problem