[Users] Updating MoL's LinearCombination to support AMR

I would like to support MoL's LinearCombination operator, which is also used to perform time integration on accelerators, to pass an additional argument "refinement level". This will help make it possible to run AMR simulations on accelerators (GPUs). This change is essentially a backport from the respective feature branch.

If not using accelerators, this change affects only MoL.

One of the main drivers for this change is to fold the CaKernel branch back into mainline. Do you agree with such experimental changes to MoL?

-erik