[Realtime][Build] Enable C++-20 for CUDA build

[1tnguyen]

Currently, CUDA source code (.cu) of realtime is compiled with C++-17 standard.

Under C++-20, <cuda/std/atomic> seems to activate the concepts/constraints code paths; hence causing nvcc to process problematic code in <type_traits> and crash.

This looks similar to the bug reported here: https://forums.developer.nvidia.com/t/compiling-a-catch2-application-with-nvcc-std-c-20-leads-a-crash-in-cudafe/255088

Reference:

I read the comment above but would like to go with C++20 here. Is it always the case that nvcc's C++20 frontend hit the internal compiler error? Or is there any workaround for the same?

Originally posted by @sacpis in #4709 (comment)