reference counted `MemoryPlot`

[arvidn]

Currently we copy the MemoryPlot objects, containing a pointer to allocated memory, and then free that memory in the destructor, unconditionally. This leads to double frees.

This is where we copy initialized MemoryPlot objects: https://github.com/Chia-Network/bladebit/blob/cuda-compression/src/tools/PlotValidator.cpp#L279