Add stream support to enable overlap between copy and compute

[mlxd]

Issue description

Description of the issue - include code snippets and screenshots here
if relevant. You may use the following template below

• Expected behavior: With the adjoint pipeline often requiring explicit memory copies between GPU buffers, as well as explicit compute calls, it can be beneficial to explicitly add support for CUDA-streams and re-enable the OpenMP threading supports. This will allow CUDA API calls that can be overlapped to do so, and hence reduce existing synchronization points. This will involves updates to the DataBuffer class, as well as explicitly to the StateVectorCudaBase and AdjointJacobianGPU classes, to better make use of the new functionality.