Enable setting file times for any writable files

[BenWiederhake]

This PR enables calling (f)utime(n)s(at) in a special way that was not possible before.

In particular, POSIX specifies that when times is a nullptr, then utimensat only requires that the caller has write access to the file. This cannot be emulated by any other function of this crate, as they always pass a userspace-provided times struct, which requires the caller to be the owner of the file.

This PR adds two new functions to the API: set_file_times_now(p: &Path, follow_symlink: bool) and set_file_handle_times_now(f: &fs::File).

• On linux/utimens/utimes, it uses the appropriate (f)utime(n)s(at) variant to pull off the trick.
  • I checked each of the three variants with both Linux 4.9 and Linux 6.6.
  • I don't have a way to check that this works as intended on aix, solaris, illumos, emscripten, freebsd, netbsd, openbsd, haiku. How do I check this?
• Android and MacOS should be posix-y enough to allow for this to work, too, but I don't feel adventurous enough to try it out. I implemented it using trivial wrappers, and included a TODO for the next person who cares to look into it.
• Redox, android, Windows use trivial wrappers, as the feature does not apply here.
• wasm lacks even the basic functionality, so I copy-pasted the "Wasm not implemented" error.

Fun fact: I'm trying to work on the Rust coreutils, and this feature is required to implement touch in a way that passes the touch/now-owned-by-other test.

[BenWiederhake]

Changes since last push:

• Don't test on Android and MacOS (obviously the times=nullptr thing won't work there because I didn't implement it!)
• Run cargo fmt

[BenWiederhake]

I ignored AT_SYMLINK_NOFOLLOW – turns out, that was a bad idea.

I'll change a few lines, but the bulk will remain identical.

[BenWiederhake]

Changes since initial version:

• Added the parameter follow_symlink: bool
• Re-ran all tests on Linux 4.9 and Linux 6.6
• Fixed a rustfmt issue
• Fixed a typo in the documentation

[BenWiederhake]

@alexcrichton Hi! Can I ask you to look into this, and perhaps merge it? This feature would be really useful.

[workingjubilee]

Android: Yes.

• https://github.com/aosp-mirror/platform_bionic/blob/d06e2e7b29ec785b9c9143b98c04dd75286fc5d5/libc/include/sys/stat.h#L178-L192

Darwin: Not immediately clear at a glance.

• https://github.com/apple-oss-distributions/xnu/blob/94d3b452840153a99b38a3a9659680b2a006908e/bsd/sys/stat.h#L582-L583
• https://github.com/apple-oss-distributions/xnu/blob/94d3b452840153a99b38a3a9659680b2a006908e/tests/utimensat.c
• https://github.com/apple-oss-distributions/xnu/blob/94d3b452840153a99b38a3a9659680b2a006908e/tests/netbsd_utimensat.c

[BenWiederhake]

Then I guess it's safest to implement it just the way that this PR already does: Defer to the non-NULL variant, and improve it in a later PR if/when more is known about what Darwin accepts.