UCT/CUDA_COPY: detect device transfers and report peak arch bandwidth

[Akshay-Venkatesh]

What

Detect if remote/local memory types for perf estimate is of type cuda/cuda-managed. If so, report peak device memory bandwidth

Why ?

Preparation for device staging pipeline protocols. Without this patch, only estimated peak host<->cuda bandwidth is reported which may not allow for device bounce buffer selection.

[Akshay-Venkatesh]

cc @yosefe @bureddy