Race condition in applications assuming IO stops after process death

[Zygmunt Krynicki]

The extcmd interface has, among others, the following two methods:
- on_line()
- on_end()

It seems natural to think that on_line() cannot occur after on_exit(). Applications making this assumption can break as that assumption is not true. If the communication pipe leaks from the started process to a third process that survives its parent's death (such as many daemons and "background" software) and the venerable process doesn't close its stdout/stderr streams then we can indeed see on_line() being called after on_exit().

To guard against this and let apps see "simplified" versions of reality. Extcmd should optionally defer on_end() and on_abnormal_end() after we're certain that no more on_line() events will occur.