Fix cppstd_flag for clang-cl driver in C++23 / C++26 / gnu++ modes

[HYUEHFJKhfjklkej]

Summary

clang-cl mimics cl.exe's /std: flag. The driver accepts only a
small set of values:

clang-cl version	/std: accepts
< 22	c++14, c++17, c++20, c++latest (+ c11/c17)
≥ 22	the above plus c++23preview

The previous mechanical flag.replace("=", ":") translation in
cppstd_flag produced /std: values that clang-cl rejects with
"argument unused during compilation" for:

• compiler.cppstd=23 on any clang-cl version (/std:c++23 is not a
  valid token);
• compiler.cppstd=26 (no /std: form exists yet);
• the pre-standard markers emitted by _cppstd_clang for older clangs:
  c++1y, c++1z, c++2a, c++2b;
• every gnu++ extension (gnu++14, gnu++17, ...).

Fix

Replace the blanket = -> : translation with a small helper that:

• emits /std:c++14, /std:c++17, /std:c++20 when clang-cl accepts
  the value directly;
• emits /std:c++23preview only on clang-cl ≥ 22 (the version that
  introduced the alias for -std=c++23);
• routes everything else through -clang:-std=... so the inner clang
  frontend receives the original GCC-style flag — the documented escape
  hatch for unsupported /std: values.

closes #19887

Test plan

The clang-cl branch in cppstd_flag was previously not covered by unit
tests. This PR adds explicit coverage:

test_clang_cl_cppstd_flags
  /std:c++14 /std:c++17 /std:c++20            (clang 17)
  -clang:-std=c++23                           (clang 17 / 21)
  /std:c++23preview                           (clang 22)
  -clang:-std=c++26                           (clang 17 and 22)
  -clang:-std=gnu++17 -clang:-std=gnu++23     (gnu++ extensions)
  -clang:-std=c++2a                           (pre-standard marker)

Suite	Result
pytest test/unittests/client/build/cpp_std_flags_test.py	12 passed
pytest test/unittests/tools/ test/unittests/client/build/	786 passed, 8 skipped (pre-existing)

Reference for the clang-cl /std: values:
https://blog.conan.io/2022/10/13/Different-flavors-Clang-compiler-Windows.html

Reference for the -clang: passthrough:
https://clang.llvm.org/docs/UsersManual.html#the-clang-cl-driver

[memsharded]

This behavior was still not approved. Ticket #19887 is labeled as "look-into" which means this is still something that the team need to consider and decide how we want to handle it. It is not that clear that the desired direction is this one, so better wait until the team checks it. Thanks.

[memsharded]

Looking good, just one clarification, but I think it can be moved forward for next 2.29

[memsharded]

As commented in #19887 (comment), I am willing to move this forward, but it would be good to align with CMake in the behavior. Is there any disadvantage to keep the -clang:-std=c++23 syntax for this combination?