On *NIX, one needs to call `waitpid()` in order for process to exit the
zombie state. If one uses `Process::isRunning()` to emulate non-blocking
wait for child process termination, process will stay zombie and function
will always return true.
This commit changes `Process::isRunning()` to call `waitpid()` with
`WNOHANG` instead of using `kill()` when checking for child process (i.e.
the one we have ProcessHandle for), which allows for process termination.
Additional trickery with mutex and event is needed to prevent exceptions
when `Process::isRunning()` and/or `Process::wait()` is called concurrently
on the same handle from different threads.
Fixes#1097.
Currently, ProcessHandle::wait (and transitively Process::wait) on *NIX
return zero if process was terminated as a result of unhandled signal.
Check if this is the case and return negative signal number instead to
indicate non-graceful process termination.
ProcessHandle handle = Process.launch(...);
handle.kill();
Then as handle gets out of scope, ~ProcessHandle would call CloseHandle() on an already closed handle.
