Time exception

[Stephen Weeks]

> I agree that making it a Word32.word is the right change, but note, you could
> still get pimped some times.  If, for instance, the interval of times spans a
> wrap, the comparison in the test would still fail.  The conclusion to  me  is
> that the test is wrong.

With this change, I don't believe Posix.ProcEnv.times will ever fail.  I'm not
sure what test you're referring to, but of course I agree that code that uses
Posix.ProcEnv.times has to take into account the possibility of wrap.
Fortunately, we're never gonna use it again, now that we have MLton.RUsage,
which delays the wrap problem for a while :-)