Copy for solver

[nHackel]

It would be helpful for solvers to provide an implementation of copy. This could allow reconstruction packages some multi-threading options.

System matrixes should generally not be deepcopied, unless the SM contains stateful information. This can be the case for certain reconstructions from MRIReco I think. Such a case could be handled with a LinearOperatorWrapper-like wrapper

[tknopp]

And the LinearOperatorWrapper-like wrapper should have immutable semantics (It won't be immutable technically but we should think of it like that)