This PR fixes a race condition in shutil.copyfile (there are several other shutil functions that should also be fixed. The current state is that only copyfile is fixed.) Unittests are passing but there are a few things that I'd like to look into:

Reading through the issue, the following questions were asked and should be looked into:

• giampaolo: I wonder whether this can introduce any change in semantic though:

  >>> import os
  >>> os.stat('foobar')
  Traceback (most recent call last):
    File "<stdin>", line 1, in <module>
  FileNotFoundError: [Errno 2] No such file or directory: 'foobar'
  >>> os.stat(333)
  Traceback (most recent call last):
    File "<stdin>", line 1, in <module>
  OSError: [Errno 9] Bad file descriptor: 333

• vstinner: Python doesn't use FD of directories by default because it introduces issues of FD limit with deep directory tree, issue of managing FD lifetime, etc. That's why some API have one flavor for path and another for FD.
• Look through the copyfile changes for operations that are duplicated/should be done inside of the openers. For instance, this PR presently stats a file twice. I think that's unnecessary.
• Look into whether there's other ways to pass the information about whether the file was created or not than via a closured variable.
• fix other shutil functions: giampaulo: All copy* functions and move() are subjects to race conditions (the only exception is rmtree()).
• Vstinner proposed that fixing copyfile is good in and of itself, regardless of whether we fix the rest of the shutil functions. I'll take a look at how similar they are before commenting on this.

This is a forward-port of #1659 from @pkmoore

• Issue: Race condition in shutil.copyfile(): source file replaced file during copy #74585