Updates for CUDA::Unified proxy

A number of changes to the CUDA::Unified proxy. These commits were originally made many months ago with the aim to improve the performance of this proxy to demonstrate the use of CUDA Unified Memory. Since this proxy is currently only used for research purposes, it does not affect any code used for 'production'.

An upcoming MR builds on top of these changes and is therefore good to have the current changes in master before they get stale (again).