statfs(2) is more involved than I expected, so this calls for its own
sample file system to test in a more focused manner.

This may bring significant performance improvements, but also comes with
some caveats. See the notes on MountConfig.DisableWritebackCaching for
details.

See the documentation on MountConfig.DisableWritebackCaching for
discussion.

We can't do this now that the kernel chooses mtimes and sends setattr
requests due to our use of writeback caching.

Otherwise we get confusing EOF errors when we later call out to osxfuse.