Consider implication of MSYS2 64-bit environments other than MINGW64

[mpilgrem]

Motivation: haskell/ghcup-hs#982

Context: https://www.msys2.org/docs/environments/

Code: Stack.Setup.setupEnv (extract):

  envRef <- liftIO $ newIORef Map.empty
  let getProcessContext' es = do
        m <- readIORef envRef
        case Map.lookup es m of
          Just eo -> pure eo
          Nothing -> do
            eo <- mkProcessContext
              ...

              $ case (sopts.skipMsys, platform) of
                  (False, Platform Cabal.I386   Cabal.Windows) ->
                    Map.insert "MSYSTEM" "MINGW32"
                  (False, Platform Cabal.X86_64 Cabal.Windows) ->
                    Map.insert "MSYSTEM" "MINGW64"
                  _ -> id

[mpilgrem]

The configuration of Stack's MSYS2 environment is implemented in version stack upgrade --source-only --git --git-branch fix6457 (#6465). The pull request includes documentation and also a new integration test. The latter sets the environment to UCRT64, uses pacman to get the UCRT64 version of gsl (mingw-w64-ucrt-x86_64-gsl), and builds an executable that depends on the Bessel function of the first kind.

It has also been tested (on 64-bit Windows) with:

• no msys-environment - uses MINGW64
• msys-environment: duff - yields a parse error
• msys-environment: MINGW32 - yields an error that the environment is inconsistent with the architecture
• msys-environment: UCRT64 - works as intended (stack path, stack config env)

Interestingly, the test seems to work also with snapshots specifying GHC < 9.6. It had been suggested that GHC < 9.6 might need a MINGW64 Stack MSYS2 environment.