functions must return `parser` not `parser.parse_args()`

Hey, thanks for a great package!! I don't think the below is a huge issue, but I just spent a few hours debugging it, so I figured I would pass it along.

In my understanding, there are two, roughly equivalent ways one might structure their argparse usage. I won't opine on the relative benefits, only that I've seen both in the wild.

method 1:

def parse_arguments():
    parser = argparse.ArgumentParser()
    # blah blah blah
    return parser

def main():
    parser = parse_arguments()
    args = parser.parse_args()

method 2:

def parse_arguments():
    parser = argparse.ArgumentParser()
    # blah blah blah
    return parser.parse_args()

def main():
    args = parse_arguments()

Method 1 works with autoprogram, and is what is outlined in the docs. However, method 2 doesn't work, and actually raises a pretty unhelpful error message. I don't think there's any reason to support method 2, and I could imagine that would be a headache to implement anyway. But, I think it might be possible to detect that a user has done this and raise a helpful error, or even simpler, add a clarifying sentence or two in the docs. (Or even simpler, just leave this issue open for others to find!)

With blessing, I'd be happy to take a stab at editing the docs accordingly. Thanks!

Start with the documentation describing autoprogram and the parse_arguments pattern. Clarify that the function must return the argparse parser rather than parser.parse_args(), and distinguish the two usage methods shown in the issue. Done means the documented requirement is unambiguous and users are less likely to encounter the unhelpful error.