RFR (S): 8237857: LogDecorations::uptimenanos is implemented incorrectly

[David Holmes]

Bug: https://bugs.openjdk.java.net/browse/JDK-8237857
Webrev: http://cr.openjdk.java.net/~dholmes/8237857/webrev/

After my changes in JDK-8235741 which changed uptimemillis to be the 
same as uptimenanos*1000000 the gtest for the logging time decorators 
started to fail intermittently on Windows where the uptimemillis values 
was not advancing sometimes after a 5ms sleep. It turned out this was 
actually a day one bug in the logging code that my change had exposed - 
the code uptimenanos was using os::elapsed_counter() but that isn't a 
time value it's just a counter value. On Linux/AIX/BSD it happens to 
count at 1ns resolution but not so on Solaris and in particular not so 
on Windows. When this counter value was truncated to "millis" there was 
no guarantee the value would increase as expected by the test.

Changes:
- use os::elapsedTimer() not os::elapsed_counter() for uptimemillis and 
uptimenanos
- remove dead code that should have been removed in JDK-8235741
- update the test by:
   - adding diagnostic printouts of the time values read
   - fix a bug where the sleep time can be too short for Windows (see 
details in bug report)

Testing:
   - gtests on all platforms with manual analysis of log files

Thanks,
David
-----