Perf: arm64: Refine disable/enable in tracecounters

Tracecounter code was not disabling counters as early as possible
and enabling them as late as possible. Refine the logic so cycles
spent inside the handler are lower. This results in more accurate
per-process counter values.

Change-Id: I5a83028da3c747c30a9e5a0ea3003638beadffec
Signed-off-by: Neil Leeder <nleeder@codeaurora.org>