Update matrix_sparse.hpp

[tpadioleau]

Update magic number

[blegouix]

Can you explain why you look at this ? It helps for the issue with OMP in Gysela ?

[tpadioleau]

Yes 1e-16 was a bit too restrictive

[tpadioleau]

As a rule of thumb I think the stopping criterion $||r||$ in double precision should follow $10^{-16} \approx \kappa(A) ||r||$, so for a good condition number $\kappa(A) < 100$, we should stop somewhere at $||r|| \in [10^{-15} , 10^{-14}]$.