Handle a race condition between process killing and restarting

If the app is killed by system_server itself, the related info could
have been cleared before binderDied comes in, so when we restart it,
we'd be unable to know if the previous instance is still alive or not.
Now keep track of this type of process kills.

Bug: 181777896
Test: Manual - Loop start camera, sleep 1, stop camera app
Test: atest CtsAppTestCases
Test: atest FrameworksServicesTests
Test: atest FrameworksMockingServicesTests
Change-Id: Icaa706b88fd40ad649defd792de1cdc05e4ae386