S390x compatibility fixes

[jchecahi]

With the new support (LLVM-18) for libomp and lld in s390x there are now tests that are unexpectedly passing (marked as failure), others that are not being run. The summary of changes in this PR is as follows:

• LIBOMP presence is detected at cmake configuration step similarly to how all the binaries are detected, using cmake's find_library.
• libomp feature added in lit.site.cfg. This is enabled depending on LIBOMP provided from cmake configuration.
• libomp related tests are now have REQUIRES: libomp lit directive. These tests will now be marked UNSUPPORTED in s390x/LLVM < 18, and will run normally on LLVM >= 18
• whole-toolchain testcases now run on s390x due to lld being supported. These tests are expected to fail, as building with -rtlib=compiler-rt will fail due to unsupported builtins library in the arch. Added XFAIL: s390x on them.

Additionally, fedora workflow in GH actions has been updated with latest supported versions.

[jchecahi]

Running this branch on a s390x system running latest Fedora Rawhide and LLVM 18.1.0~rc4 shows no failures:

# lit --show-xfail --show-unsupported -v tests
-- Testing: 51 tests, 2 workers --      
PASS: LLVM regression suite :: plugins_newpm.txt (1 of 51)
[...]
PASS: LLVM regression suite :: scanbuild_py.cpp (51 of 51)
********************
Unsupported Tests (1):
  LLVM regression suite :: test_hwasan.c

********************
Expectedly Failed Tests (5):
  LLVM regression suite :: lto_and_cfi.c
  LLVM regression suite :: lto_and_cross_dso_cfi.c
  LLVM regression suite :: unwind.cpp
  LLVM regression suite :: whole-toolchain.c
  LLVM regression suite :: whole-toolchain.cpp


Testing Time: 12.03s

Total Discovered Tests: 51
  Unsupported      :  1 (1.96%)
  Passed           : 45 (88.24%)
  Expectedly Failed:  5 (9.80%)

[jchecahi]

@sylvestre Could you take a look at the cmake logic and double check if it's a good way to approach the libomp detection?
Also, it seems that the ubuntu runs that failed are unrelated to the changes in this PR. They've been there for a while.

[jchecahi]

@nikic pointed out internally that detecting libomp using clang -fopenmp could mask real failures in a scenario where clang search path is broken, for example. I updated the PR using a different approach based on finding the library instead.

[jchecahi]

@sylvestre ping, pretty please? :)

[tuliom]

While I'm not a maintainer of this project, I did take a look at this PR.
With this PR, no tests fail on s390x.
Nice catch fixing the missing documentation!

The tests failing on Ubuntu are unrelated to this PR.

[sylvestre]

sorry, i missed your notif?!