[FEA]: Implement a memory_resource using cudaMalloc and cudaFree

[miscco]

In order to move our <memory_resource> plans forward we want to implement a memory resource that utilizes cudaMalloc and cudaFree to allocate memory.

That memory_resource should live in <cuda/memory_resource> and at least satisfy the cuda::mr::resource_with<cuda::mr::device_accessible> requirement.

Poll for the audience. Should it also satisfy cuda::mr::async_resource_with<cuda::mr::device_accessible>?

[miscco]

I have for now removed the async interface. We do want to add a cuda_async_memory_resource anyway, so adding an essentially false promise seems bad