AthenaExternals CUDA Setup, master branch (2021.11.19.)

Made AthenaExternals set up CUDA from an LCG release automatically, if it is available. Thereby not relying on AtlasSetup to do so. This is in connection to the issue discussed in ATLINFR-4342.

As it turns out, modifying environment variable through ENV{PATH} and friends, is actually enough to even make it possible to build CUDA code "natively" as part of AthenaExternals. I tested this by adding some dummy code to the Gdb package, like the following:

enable_language(CUDA)
atlas_add_executable( gdbCudaTest src/cudaTest.cu )

And that actually worked... In an environment in which CUDA was not set up at all prior to calling cmake.

So I'm now starting to think that as long as we'll be getting CUDA from our LCG releases, we can pretty much stop setting it up with AtlasSetup. We'll need to re-visit this if at one point we'll want to make use of CUDA in AnalysisBase. But that will not happen for a while. (Or ever...)