krylov method always check for l2 norm, slowing down

[AnderBiguri]

This should be a user option, e.g. https://github.com/CERN/TIGRE/blob/master/Python/tigre/algorithms/krylov_subspace_algorithms.py#L66
or
https://github.com/CERN/TIGRE/blob/master/MATLAB/Algorithms/CGLS.m#L94

We should have them on by default, but allow the user to disable it.