fastcolumn, fastrow become ineffective after to_cpu, to_gpu

[Shayan-P]

In the current implementation, to_cpu, to_gpu copy the whole data. This means that Adjoint matrices get copied thus they are not adjoint anymore.

As a result, the user has to reapply fast_row, and fast_column whenever copies/collects data.

This can be fixed by dispatching a different to_cpu/to_gpu function for Adjoint types.