Inconsistent ParseResults from parse actions that return other types

[djpohly]

The explanation of parse action handlers in #309 gives three cases for the return type:

• None: use the existing results object, potentially modified by the action
• ParseResults: replaces the existing results object
• Anything else: replaces the existing results object, wrapped in a ParseResults

In the "anything else" category, many types work as described. Notice in each case that there is a single element in the tokens list, which is identical to results["letters"]:

>>> Empty()("name").set_parse_action(lambda: "123").parse_string("")
ParseResults(['123'], {'name': '123'})
>>> Empty()("name").set_parse_action(lambda: {1,2,3}).parse_string("")
ParseResults([{1, 2, 3}], {'name': {1, 2, 3}})

If the action returns a tuple, the token list is correct, but the names dict is wrong:

>>> Empty()("name").set_parse_action(lambda: (1,2,3)).parse_string("")
ParseResults([(1, 2, 3)], {'name': 1})
  should be: [(1, 2, 3)], {'name': (1, 2, 3)}

If the action returns a list, both the tokens list and names dict are wrong:

>>> Empty()("name").set_parse_action(lambda: [1,2,3]).parse_string("")
ParseResults([1, 2, 3], {'name': 1})
  should be: [[1, 2, 3]], {'name': [1, 2, 3]}

You can even get some very unexpected behavior from other types:

>>> Empty()("name").set_parse_action(lambda: collections.defaultdict(str)).parse_string("")
ParseResults([defaultdict(<class 'str'>, {0: ''})], {'name': ''})
  should be: [defaultdict(<class 'str'>, {})], {'name': defaultdict(<class 'str'>, {})}