Add clang-16 test runner

[j-stephan]

Fixes #1937.

[j-stephan]

Looks like clang-16 is generating wrong PTX code:

ptxas /tmp/counterBasedRng-82eedc/counterBasedRng-sm_61.s, line 10521; fatal   : Parsing error near '.': syntax error
ptxas fatal   : Ptx assembly aborted due to errors
clang: error: ptxas command failed with exit code 255 (use -v to see invocation)

[bernhardmgruber]

@j-stephan, would you have some time to bring your changeset in? icpx was updated now and also shows these errors in tinymt:

/home/runner/work/alpaka/alpaka/include/alpaka/rand/TinyMT/tinymt32.h:102:9: error: unsafe buffer access [-Werror,-Wunsafe-buffer-usage]
    y = random->status[3];
        ^~~~~~~~~~~~~~

that this PR is fixing, and it's blocking other PRs now as well, for example: #1988. Thx!

[SimeonEhrig]

This PR fixes the readthedocs problem: #2010

[SimeonEhrig]

Actual we should merge this PR before #2005. The PR fixes also a bug in the generator script, which causes that no Clang CUDA runtime test are generated.

[SimeonEhrig]

I recommend to rebase the PR on dev and rerun the CI with filter rule: CI_FILTER: ^linux_clang-cuda16

Maybe other version than CUDA 11.3 does not work and we don't see it, because the jobs was in a later wave. With the filter rule, only 5 jobs are generated.

[j-stephan]

There is an open bug report over on the LLVM issue tracker: llvm/llvm-project#58491 I will look into the work-around tomorrow.

[SimeonEhrig]

you need to update the requirements.txt of the job generator