(10) RFR: 8157709: NMT should use size_t version of Atomic::add

[David Holmes]

Bug: https://bugs.openjdk.java.net/browse/JDK-8157709

webrev: http://cr.openjdk.java.net/~dholmes/8157709/webrev/

NMT typdef'd MemoryCounterType to jlong on 64-bit and jint on 32-bit. 
Really it should just be size_t and in fact the typedef is not needed at 
all as it was only used for casting size_t's when invoking Atomic::add. 
So MemoryCounterType is removed. Only glitch was that using negation of 
an unsigned variable triggered a warning on Windows which I have had to 
silence with a pragma. (As unknown pragma's must be ignored I have not 
ifdef'd the pragma - doing so would also be a problem as warning 
suppression only operates on the next line, so then I'd have to push and 
pop the warning which would be really, really ugly.)


Also as a result of this change there is no longer any use of the buggy 
Atomic::add(jlong) and as discussed here:

http://mail.openjdk.java.net/pipermail/hotspot-dev/2016-February/021620.html

it is simply removed.

Testing: JPRT
          runtime/NMT

Thanks,
David