add flag harakiri-only-graceful

[atwrg]

this setting can be useful for cases when graceful harakiri signal is handled by a function in app and from that function is raised exception and we don't want to reload the worker using effective harakiri, currently as a result of incrementation pending_harakiri variable, first harakiri on the given worker is graceful and second is effective harakiri with reload

example function in app:

def exit_gracefully(*args):
    raise TimeoutError

signal.signal(signal.SIGSYS, exit_gracefully)