Perf: Make per-process counters cumulative

In order to get overall value for performance counters, remove reset during
context switches. Add mechanism to calculate deltas for output to the trace
buffer.

Change-Id: I03d894a1b3bd400d8c256c3068f917ed001fbfd9
Signed-off-by: Sheetal Sahasrabudhe <sheetals@codeaurora.org>