slepc solver default threshold should be reduced

[sangallidavide]

As discussed with some of the developers, the default threshold for the slepc solver is too low for a single precision compilation
BSS_slepc_tol=1E-6_SP

With slepc convergence stops when
norm(Hx-lambdax)/(abs(lambda)*norm(x)) < tol

The default is slepc library in SP is 1.E-5. I propose to adopt the same default.
This would roughly guarantee 5 correct digits in both the eigenvalue and the eigenvector.

[attacc]

Fine for me