Memory allocation of NrnThread with CudaMallocManaged / posix_memalign

[pramodk]

We have changed allocation of NrnThread with new. There were two use cases:

• when we use aligned pragma with intel compilers we needed to have nt to be aligned
• for unified memory with cuda, we wanted it to be ideally cudaMallocManaged

This would be good to have some way?

@alkino did PR to implement new memory allocator which should be improved to incorporate above use cases.