cuml: remove cuml solver as it can not fit the current framework (tolerance and maxiter are hardcoded), also the current version could not be run (that's why we didn't put cuml in the paper to begin with for the ridge regression)
glmnet: add a docstring warning and skip test for now (cf #2)
Modifications on solvers