gh-101357: use os.path.exists() etc from pathlib

[barneygale]

The following pathlib.Path methods now directly call functions in os.path:

• samefile()
• exists()
• is_dir()
• is_file()
• is_symlink()

In conjunction with #101324, this improves performance of these methods by using making fewer system calls than stat()

The original generic implementations of these methods are preserved in a new, internal _PathWithStat class. In future this may form the basis of a public AbstractPath class. The generic implementation of is_mount(), which was removed in 29650fea, is also restored to this class.

---

• Initial implementation
• Wait for gh-101196: Make isdir/isfile/exists faster on Windows #101324 to land
  • And also a new strict argument to the functions?
• Measure performance
• Write news blurb

• Issue: Improve Windows performance of pathlib.Path.is_file and friends #101357

[mdboom]

#101324 has landed. The general approach here makes sense. I imagine what we'll want to do is:

• add a strict flag to the C functions
• add a Python wrapper around them for os.path (since a new kwarg would be an API change that we probably want to avoid
• use them from here

Alternatively, we could make the C functions always raise the exceptions that pathlib raises, and the Python wrappers in os.path would catch those exceptions and return False. I think microbenchmarking is the only way to know if any of that matters. I think we will lose a little bit of the gains of #101324 either way, but that's ok as long as it's not significant, I guess.

[barneygale]

since a new kwarg would be an API change that we probably want to avoid

We added a similar strict keyword-only argument to os.path.realpath() recently - couldn't we do that here too?

[barneygale]

I opened this PR mainly to demonstrate how to preserve pathlib's stat() helpers, but I don't have the time to bring it forward. Closing so that someone else might have a go.