UR's UR_KERNEL_INFO_NUM_ARGS query is size_t according to the spec, but this is a little strange given its closest equivalent CL_KERNEL_NUM_ARGS is cl_uint. If we were to align with CL on this the CL implementation would be a little bit nicer for being able to skip the cast, and anything further up the stack that might expect us to align with CL would no longer potentially be surprised.