This commit contains few bugfixes:
 - The Taylor expansion for imaginary time was not correct.
 - The code was wrongly mixing the density between timesteps of the evolution
 - The code was doing a subspace diagonalization which is not correct
 - The orthonormalization routine was not stable and is replaced by the standard orthogonalization.
 - The code was freezing states once converged, which is not correct unless independent particles

Besides this, the code now check that the exponential is stable. The default time-step is set to a valid
close to the one shown in the litterature.