Thanks to some hints by eric.smith, I've run the regression tests with
this patch applied. It turns out there are still some outstanding issues
with draft 9.

One issue was in test_glob.py where broken symlinks would fail to be
matched by glob. This was due to the fact that lexists still needed to
be implemented in ntpath.py. Draft 10 includes this capability.

Another problem the regression tests elicited is more insidious. The
patch breaks test_os.test_1686475, which references issue1686475. The
new symlink-aware stat must traverse symlinks. Unfortunately, the
current method of resolving the symlink target (as recommended by
Microsoft) fails to work if the symlink target is in use (or otherwise a
handle cannot be obtained). Unfortunately, it's not possible to call
GetFinalPathNameByHandle without a handle, the same way as issue1686475
was addressed by using a different API call.

I'm still devising a workaround for this undesirable behavior, but I
wanted to report that progress is still being made and I wanted to
capture the intermediate patch for posterity.

Stay tuned for more info.