Fix U+2010 typo which prevents successful building on MSYS2 CLANGARM64

[GulinSS]

GHC JS Backend does not support TemplateHaskell and Windows. To investigate this issue I was needed first to build GHC itself to run on Windows 11 ARM (because this hardware is available for me). It was done here. And it allowed me to dive in the main issue.

This change allows running a build of GHC JS Backend cross-crompiler by using Windows 11 ARM and a non-cross GHC build for Aarch64.

Without this change LLVM compilers at Windows are unable process the build.

[hasufell]

How do we test this?

[GulinSS]

1. Take Aarch64 hardware. I use Apple aarch64 with UTM and Windows for ARM 11 installed. It works surprisingly fine.
2. Use fresh bindist ghc-9.12.1-aarch64-unknown-mingw32.tar.xz as a bootstrap compiler (you can also download the archive, move it to more stable place and add to ghcup metadata).
3. Install MSYS2 with CLANGARM64 support. LLVM would be installed already.
4. Install git-for-windows (or pacman -S git).
5. Take cabal https://github.com/haskell/ghcup-metadata/blob/develop/ghcup-0.0.9.yaml#L6261. X86 is ok for it.
6. Prepare your build environment (you will need at least happy, alex, python, nodejs, emscripten).
7. Patch cabal files of built-in libraries to work with ghc-12.1 (it should support ghc-13...)
8. Run build of `ghc js backend.

At stage1 your will meet the error about this strange symbol which I removed from sources by this PR.

Also, to build stage0 you will need apply these patches to built-in libs: https://gitlab.haskell.org/ghc/ghc/-/issues/24603#note_602554

---

I am not sure does it present at x86 windows but I think that LLVM usage at Windows is somehow related to this issue because earlier GCC was used instead.

[hasufell]

Well, I'm not sure I have time for that. And it's hard for unix to maintain aarch64 windows compatibility if there is no CI that tells us when we break things.

[GulinSS]

Depends on requested changes. It looks like a typo which breaks compiling stage. Somebody wanted to use - but substituted something from UTF. 🤔

It was ‐ (https://www.compart.com/en/unicode/U+2010) not just - (https://www.compart.com/en/unicode/U+002D). I just removed it because it looked like a typo.

---

I have no x86 hardware around. You could try to build GHC JS Backend by usual way at Windows but use LLVM as a compiler instead of GCC. May be it could help to reduce lots of steps from my origin message and save your time on reproduction.

---

I've renamed the PR to reflect better sense of these changes.