Make a time test less flaky

This test was failing periodically on windows and other platforms, and in
debugging the issue locally I've found that the previous test was failing
at the assertion `ns0 <= ns1`. Upon inspecting the values, the two numbers were
very close to one another, but off by a little bit.

I believe that this is because `precise_time_s` goes from `u64` -> `f64` and
then we go again back to `u64` for the assertion. This conversion is a lossy one
that's not always guaranteed to succeed, so instead I've changed the test to
only compare against u64 instances.

Author: alexcrichton

src/libextra/time.rs

@@ -1005,18 +1005,17 @@ mod tests {
 
     fn test_precise_time() {
         let s0 = precise_time_s();
-        let ns1 = precise_time_ns();
-
         debug!("s0={} sec", f64::to_str_digits(s0, 9u));
         assert!(s0 > 0.);
-        let ns0 = (s0 * 1000000000.) as u64;
-        debug!("ns0={:?} ns", ns0);
 
-        debug!("ns1={:?} ns", ns0);
+        let ns0 = precise_time_ns();
+        let ns1 = precise_time_ns();
+        debug!("ns0={:?} ns", ns0);
+        debug!("ns1={:?} ns", ns1);
         assert!(ns1 >= ns0);
 
         let ns2 = precise_time_ns();
-        debug!("ns2={:?} ns", ns0);
+        debug!("ns2={:?} ns", ns2);
         assert!(ns2 >= ns1);
     }