Make switch user test measurements consistent with each other.

In UserLifecycleTests switchUser and switchUser_running test methods
wait for UserSwitchObserver.onUserSwitchComplete callback,
whereas switchUser_stopped waits for ACTION_USER_UNLOCKED broadcast
after triggering the user switch.

Which makes the results incomparable and also shows switchUser_stopped
is running faster than switchUser_running, but it makes more sense
the other way around.

This CL makes sure the tests wait for and measure the same thing
to make the results comparable.

Bug: 242843357
Test: atest MultiUserPerfTests:android.multiuser.UserLifecycleTests#switchUser_stopped --no-bazel-mode
Change-Id: I250ceef1ae9ce0adec40972b7962932066121c4f