Re: Final(?) proposal for wal_sync_method changes

Given my concerns around exactly what is going on in the Windows code,
I'm now afraid to mess with an all-platforms change to fdatasync as the
preferred default; if we do that it should probably just be in HEAD not
the back branches.  So I've come around to the idea that Marti's
proposal of a PLATFORM_DEFAULT_SYNC_METHOD symbol is the best way.
(One reason for adopting that rather than some other way is that it
seems quite likely we'll end up needing it for Windows.)

I haven't touched the documentation yet, but attached is a proposed
code patch against HEAD.  This forces the default to fdatasync on Linux,
and makes some cosmetic cleanups around the HAVE_FSYNC_WRITETHROUGH_ONLY
confusion.

			regards, tom lane