[clang] [clang] Support --sysroot= for ${arch}-windows-msvc targets (PR #96417)

[Reid Kleckner via cfe-commits]

rnk wrote:

Please avoid derailing into off-topic conversations about hardware/software freedom and control. All these things are important to me (even as a Google employee), but having to read through them to find principled objections to the patch as-is is an obstacle to approving it!

It seem to me the point of disagreement is that @sharadhr and @zmodem want the new `--sysroot` codepath to share logic with the main codepath, which examines environment variables and respects `/winsysroot`. First, is that correct?

Second, why do folks want to share these codepaths? It is difficult to support mixing Unix and Windows header and library layout conventions. If the user has something that looks like a sysroot as used by the Unix world, it seems reasonable to me to have that bypass the normal Windows SDK detection logic.

I think my next concern is the way we model the Microsoft C++ standard library. That seems to have some subtlety, and I'm unsure if I like assigning the name "stl" to the microsoft STL, but I could go with it. It's kind of [github-first](https://github.com/microsoft/stl). I guess I see Hans suggested "msstl" too. If I had to pick one, I'd prefer that one. For a reader, `-stdlib=stl` doesn't feel like it provides enough information.

There are many, many ways to build software, and instead of arguing which one is right, I'd like to focus on principled objections against this approach to building software for Windows using Unix conventions.

https://github.com/llvm/llvm-project/pull/96417