Load tracking cpu migration improvements

Created by: qais-yousef

The 2 patches make the {One,Two}Task[s]CPUMigrationTest more stable by reducing some sources of noise that made the tests fail occasionally at some iterations and during one integration test (20180720) fail 100% of the iterations.

I can't root cause why 20180720 integration was worse than others but changes to NFS might have caused more kworker threads to run randomly on the CPUs the test was running on causing utilization to be slightly off. Combined with rounding/quantization errors in calculating util_mean in _get_util_mean() they all were adding up enough to tip the point to a 100% failure rate.

Depending on the next integration test results with these fixes - more investigation might be necessary.