I don't get why the new spec fails on Aarch64 CI and in the interpreter only. Aarch64 CI is the runs-on environment which uses a different base image than the hosted runners from GitHub. But interpreter specs should run in exactly the same environment as other jobs. 🤷
Even the guard to make sure the file isn't readable didn't help. The file has no permissions. Yet still, apparently the file handler can still read its contents. This is very weird.

At last, I added a File.read test in order to properly detect whether the file is actually readable. If anyone has any better idea, I'm happy for suggestions. Maybe we can introduce some other kind of error?

broken-symlink.txt is breaking Git on MSYS2, it should be created and deleted entirely within the new specs:

$ MSYS='winsymlinks:nativestrict' git restore --source=HEAD :/
error: unable to create symlink spec/std/data/static_file_handler/broken-symlink.txt: No such file or directory

(MSYS='winsymlinks:native' would successfully create an invalid WSL1 symlink using IO_REPARSE_TAG_LX_SYMLINK, which we don't want either.)

Also, why does the spec look for broken_symlink.txt with an underscore instead of a hyphen? Is this intentional?