RFR (S): 8174734: Safepoint sync time did not increase

[David Holmes]

Bug: https://bugs.openjdk.java.net/browse/JDK-8174734
webrev: http://cr.openjdk.java.net/~dholmes/8174734/webrev/

This is a simple logic flaw in a test that expected to see the safepoint 
sync time always increase but that's not the case because it only has 
millisecond resolution and only measures the time taken to reach a 
safepoint (not the time consumed by the safepoint)..

This was originally fixed by Coleen:

http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/2018-February/026138.html

but I was confused as to what was actually happening and so took it 
over. I decided to rewrite the test to actually check what can be relied 
upon:
  - safepoint counts and times are always positive
  - safepoint counts are accurate (at least as many as expected)
  - safepoint time never goes backwards

Testing:
   - running the test 50 times on each platform (across a range of 
machines) to see if we've ironed out the intermittent failures

Thanks,
David