[FEA] Support PTX view in Devtools (e.g. NSight Compute)

[gmarkall]

numba-cuda presently supports mapping the SASS back to Python source code via the lineinfo kwarg.

However, this only enables correlation of the SASS and Python source. The PTX generated during the compilation process does not appear to be visible to NSight Compute.

In CUDA C / C++, it is possible to view the PTX alongside the source code. This request is to enable similar functionality for Numba-CUDA.

Screenshot shows equivalent functionality to that requested, for a C++ kernel. Created with

nvcc easier.cu -lineinfo
ncu --set=full --import-source=on -o easier.ncu-rep ./a.out

from https://developer.nvidia.com/blog/even-easier-introduction-cuda/

cc @lucas-erlandson-nvidia