NEWHPROF: Convert to use JVMTI 0.3.16 GetThreadListStackTraces()

This is a performance issue for cpu=samples that has to sample all thread
stacks every 'interval' ms (default 10, see interval option: java -Xrunhprof:help).

Current loop over GetStackTrace() for N threads does N safepoints, stopping
all threads N times, this does 1 safepoint, stopping threads once.