To avoid long intervals between taking samples due to processing all accumulated samples at once, the samples are processed one by one and we check if the sampling interval has elapsed after each step rather than after processing all the samples in the queue.

This is a modified version of r16549 whith a fix for test flakiness. The test flakiness introduced by the previous version of this changed was fixed by changing return type of ProfilerEventsProcessor::ProcessOneSample from bool to enum with 3 options. In the main profiling loop we decide that the next code event should be processed when sample with a greater ordinal number is encountered. When processing remaining samples we shouldn't wait for more samples and if the samples queue is empty we just process next code event.

BUG=v8:2814,v8:2871
R=bmeurer@chromium.org, loislo@chromium.org

Review URL: https://codereview.chromium.org/23455036

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16564 ce2b1a6d-e550-0410-aec6-3dcde31c8c00