When a new GC cycle starts, the bytes_allocated_since_last_gc() is reset to zero. The first allocation rate sample after completion of GC typically sees a decrease in the value of bytes_allocated_since_last_gc(),which causes it to ignore all allocations that occur during concurrent GC.

Fix the implementation of ShenandoahAllocationRate::sample() so that these allocations are not ignored from computation of average allocation rate.