[Bug] Review GPSR with degenerate spectrum

[jpmoutinho]

As mentioned by @vytautas-a:

I made some calculations and experiments related to this problem (in context of GPSR) and it seems that it is two-fold: one related to numerical stability of linear system solver and another more fundamental and related to the spectral structure of some generators. For example, G=4*X(0)+2.001*X(1) is handled perfectly, however G=4*X(0)+2*X(1) - fails. The first case has 4 distinct spectral gaps, 2 of which are very close to 8, the other - only three. It seems that such degeneracy in eigenspectrum breaks down the method because the numerical solver has to deal with linear equation system that has certain symmetry.