Try to fix *sum on windows

[RenjiSann]

Attempt to fix #6655

[github-actions]

GNU testsuite comparison:

Congrats! The gnu test tests/basenc/bounded-memory is now passing!

[RenjiSann]

@YDX-2147483647 @wu0lss4j Could you test with this branch if that fixes your problem ?

[YDX-2147483647]

Thank you for working on it!

Here're the results for a.fake-ttf + this PR on my machine.

❯ cargo r -r -- sha256sum a.fake-ttf
1c671d7322d49cd2726475f4b8a8b50f27b454789e23a31c6ac14014740d8e58  a.fake-ttf
❯ cargo r -r -- sha256sum --text a.fake-ttf
1c671d7322d49cd2726475f4b8a8b50f27b454789e23a31c6ac14014740d8e58  a.fake-ttf
❯ cargo r -r -- sha256sum --binary a.fake-ttf
1c671d7322d49cd2726475f4b8a8b50f27b454789e23a31c6ac14014740d8e58 *a.fake-ttf

❯ cargo r -r -- sha256sum --text a.fake-ttf | cargo r -r -- sha256sum --check
a.fake-ttf: OK
❯ cargo r -r -- sha256sum --binary a.fake-ttf | cargo r -r -- sha256sum --check
a.fake-ttf: OK

This PR fixes the problem of sha256sum --check, but also introduces unwanted changes.
Compare this PR with 0.5.0:

❯ coreutils --version
coreutils 0.5.0 (multi-call binary)
❯ coreutils sha256sum a.fake-ttf
1c671d7322d49cd2726475f4b8a8b50f27b454789e23a31c6ac14014740d8e58 *a.fake-ttf # 👈 I don't think `*` should be removed
❯ coreutils sha256sum --text a.fake-ttf
73c0a22f7ed1c54ebde921712ced33d9458975cd86e9fe4757257fdd6293e512  a.fake-ttf # 👈 I think this is the correct output, and should not be changed.
❯ coreutils sha256sum --binary a.fake-ttf
1c671d7322d49cd2726475f4b8a8b50f27b454789e23a31c6ac14014740d8e58 *a.fake-ttf # Unchanged and okay

Edit: Should the following fail?

❯ coreutils sha256sum --text a.fake-ttf | cargo r -r sha256sum --check
a.fake-ttf: FAILED
sha256sum: WARNING: 1 computed checksum did NOT match

[RenjiSann]

This PR is not the reason for this unwanted behavior, sicne I only changed a piece of code that is run in --check mode only. I'll investigate

[RenjiSann]

Actually, that reopens the debate of --text and --binary handling #9168.

The "unwanted" behavior for cksum -a sha256 a.ttf is the one actually expected if we follow the GNU documentation

 ‘--binary’
...
This option is the default on systems like
MS-DOS that distinguish between binary and text files

At the time, we decided to ignore --binary as it didn't change the behavior on UNIX and GNU didn't work on Windows, but now I realize that we should still be coherent at least for the --check option.

[github-actions]

GNU testsuite comparison:

GNU test failed: tests/env/env-signal-handler. tests/env/env-signal-handler is passing on 'main'. Maybe you have to rebase?