"Might have been killed, might have crashed" is not helpful, and the inability to know is the very issue at hand here. You can't know whether your application crashed, the user forcefully killed it or you fell afoul a background killer. What now, do you start spamming the user anyway?
Most of the time when you crash there is a signal associated with that which can be handled. If you application terminated normally under an error case, then that has to be captured through the same mechanism.
"Might have been killed, might have crashed" is not helpful, and the inability to know is the very issue at hand here. You can't know whether your application crashed, the user forcefully killed it or you fell afoul a background killer. What now, do you start spamming the user anyway?