Support multicasting functionality

[eyalroz]

Beginning with CUDA 12.2, the driver introduced support for multicasting of memory operations. This involves the following API calls as of CUDA 13.1:

CUresult cuMulticastCreate(CUmemGenericAllocationHandle *mcHandle, const CUmulticastObjectProp *prop);
CUresult cuMulticastAddDevice(CUmemGenericAllocationHandle mcHandle, CUdevice dev);
CUresult cuMulticastBindMem(CUmemGenericAllocationHandle mcHandle, size_t mcOffset, CUmemGenericAllocationHandle memHandle, size_t memOffset, size_t size, unsigned long long flags);
CUresult cuMulticastBindMem_v2(CUmemGenericAllocationHandle mcHandle, CUdevice dev, size_t mcOffset, CUmemGenericAllocationHandle memHandle, size_t memOffset, size_t size, unsigned long long flags);
CUresult cuMulticastBindAddr(CUmemGenericAllocationHandle mcHandle, size_t mcOffset, CUdeviceptr memptr, size_t size, unsigned long long flags);
CUresult cuMulticastBindAddr_v2(CUmemGenericAllocationHandle mcHandle, CUdevice dev, size_t mcOffset, CUdeviceptr memptr, size_t size, unsigned long long flags);
CUresult cuMulticastUnbind(CUmemGenericAllocationHandle mcHandle, CUdevice dev, size_t mcOffset, size_t size);
CUresult cuMulticastGetGranularity(size_t *granularity, const CUmulticastObjectProp *prop, CUmulticastGranularity_flags option);

we should support this functionality.