strftime and timezone patch patched

[DJ Delorie]

But I *don't* want cygwin_tzset to have a chance to define TZ.  I want
the new code in tzsetwall() to do it.  If cygwin_tzset() needs to be
called again *after* TZ is defined, fine, but the code in
cygwin_tzset() that sets TZ from the NT information is woefully
inadequate.  I added code from a different contributor that makes
tzsetwall() do a more complete job of setting TZ from the NT
information - that should be used instead, and thus cygwin_tzset needs
to be called further down in the code (after the call to tzsetwall()
at least).

> 	* localtime.c: calls cygwin_tzset again.
> 
> -#if 0 /* we set it more accurately in tzsetwall() */
> +#if 1 /* we set it more accurately in tzsetwall() */