The implementation from libc++ is not complete. It lacks
support for static_simd_cast, which we use. For more information,
please see https://libcxx.llvm.org/Status/Parallelism.html

Notes:
 - Cannot enable C++20, std::simd headers are not ready for nvc++
 - Cannot enable Vc or Umesimd backends, as they don't support nvc++

This causes undefined behavior with GCC 15 and UMESIMD,
and compilation failures with nvc++ (no support for -ffast-math).

Fix errors due to -Wmissing-template-arg-list-after-template-kw.

GCC 15 is the first compiler to notice this problem:

In file included from VecCore/include/VecCore/VecCore:16,
                 from VecCore/test/test.h:6,
                 from VecCore/test/interface.cc:1:
VecCore/include/VecCore/Backend/ScalarWrapper.h:
VecCore/Backend/ScalarWrapper.h:157:57: error:
 'const class vecCore::WrappedScalar< <template-parameter-1-1> >' has no member named 'val_ptr' [-Wtemplate-body]
  157 |   WrappedScalar(const WrappedScalar *const s) : fVal(s->val_ptr) {}
      |                                                         ^~~~~~~

This enables and tests the std::simd backend for testing.

The script can be generically called as

ctest -VV -S cmake/TestVecCore.cmake

from the top directory of the repository. There are several options
to customize the build, the main ones are:

-DCOVERAGE=1         Enables test coverage analysis with gcov
-DMEMCHECK=1         Enables memory checking with valgrind
-DSTATIC_ANALYSIS=1  Enables static analysis with clang-tidy
-DINSTALL=1          Enables an extra step to call make install

When enabling coverage, a report is generated by default in the
coverage/ directory inside the build directory. The results can be
viewed by opening the file coverage/coverage_details.html. This step
can be disabled by passing -DGCOVR=0 to ctest. It is recommended to
use a build with debug info and low optimizations to generate the
coverage analysis.

The configuration can also be specified directly on the command line
via the -C option. For example, to run a coverage build with less verbose
output, and using AVX2 SIMD ISA, run:

ctest -V -C RelWithDebInfo -DCOVERAGE=1 -S cmake/TestVecCore.cmake,AVX2

Some environment variables can also influence the behavior of the
script, like CC, CXX, CMAKE_GENERATOR, CTEST_CONFIGURATION_TYPE,
CMAKE_BUILD_PARALLEL_LEVEL, CTEST_PARALLEL_LEVEL, and CMAKE_ARGS.
These are mostly self-explanatory and can be used to override the
provided defaults. For example, to build with the clang compiler
and use Ninja as CMake generator, one can run:

env CC=clang CXX=clang++ CMAKE_GENERATOR=Ninja ctest -V -S cmake/TestVecCore.cmake

Finally, the script tries to load configuration files from the .ci
subdirectory in the source directory. The default configuration is
used if no specific configuration is found for the detected OS. For
example, on Ubuntu, a file named ubuntu.cmake will be used if present
instead of config.cmake. The script also tries to detect a version,
so, for example, on Alma, one could use almalinux8.cmake which would
have higher precedence than almalinux.cmake.