Use pool memory resource for CUDA allocations

[mlxd]

Issue description

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

With batching of observables in the adjoint GPU pipeline, often multiple allocations and frees happen for state-vectors, which can create a natural synchronization point in the GPU, as well as introducing delays for the allocations and frees. Having a preallocated pool resource that is used to enable resuse of statevector memory blocks would remove the need to perform explicit frees, when the memory resource can simply be reset and reused. For the above, I proposed either adapting the DataBuffer class, and the DeviceResource class to track a fixed number of GPU buffers, or to provide explicit bindings to RMM's pool memory resource.