bug#21020: 24.4; `display-time-world' tampers with TZ

[Paul Eggert]

Eli Zaretskii wrote:

> > test/automated/icalender-tests.el exercises the new code.
>
> But the tests there that play with TZ are all expected failures on
> MS-Windows, because they use Posix format of time-zone definition that
> Windows runtime doesn't support.  So that probably means there are no
> tests that could be used on Windows.

OK, didn't know that.  Presumably this could be addressed by adding tests for 
MS-Windows TZ settings, if somebody has the time for this (pun intended...).

> > Also, the MS-Windows port can't entirely skip the time module any
> > more, as that module arranges for time.h to declare localtime_rz and
> > friends.
>
> This could conflict with time.h in nt/inc/sys/, no?

I don't see why.  <sys/time.h> is a different API.  This patch affects only 
<time.h>.

> lib/time.h seems to want to replace 'localtime', which might
> clash with 'sys_localtime' we currently use as replacement on w32.

The replacement is conditional, and the condition should be false on w32.