Add definition of sycl-nvptx64-nvidia-cuda pass

[Pennycook]

Related issues

• I somehow missed this during Add compiler definitions #168.

Proposed changes

• Add a regression test using -fsycl-targets=nvptx64-nvidia-cuda.
• Define the missing sycl-nvptx64-nvidia-cuda pass.

[laserkelvin]

LGTM - is there an exhaustive set of fsycl-targets documented somewhere? I did a quick search and this seems to be it

[Pennycook]

LGTM - is there an exhaustive set of fsycl-targets documented somewhere? I did a quick search and this seems to be it

I don't think so. I think the idea is that other plugins or LLVM variants could add their own targets here.

Users should be able to work around that by defining custom compiler behavior (i.e., defining a new pass for icpx associated with the right mode). But because compiling SYCL for NVIDIA is so common, I think it makes sense for us to include the definition by default.