ENH: slightly prefer split before named assigns

[apbard]

Following up #691, this is an attempt to fix it by slightly favour split before named assign.
In the current state it only mitigates the effect in some cases:

# now ok
def fooooooooooooo(aka=['aaaaaaaa', 'aaaaaaa', 'aaaa', 'aa', 'a'],
                   types=['double', 'bool'],
                   validator=some_looooooooooong_name):
    pass

# still bad
def fooooooooooooo(type01=['double'], type02=['double'], type03=['double'],
                   aka=['aaaaaaaa', 'aaaaaaa', 'aaaa', 'aa',
                        'a'], AAAAA=['double', 'bool'],
                   validator=some_looooooooooong_name):
    pass

Also it breaks a couple of tests that are right the way they are now. Example:

  def f():
    if True:
      pytree_utils.InsertNodesBefore(
          _CreateCommentsFromPrefix(
              comment_prefix, comment_lineno, comment_column,
              standalone=True), ancestor_at_indent)  

@gwelymernans Any idea where/how to fix this?

[bwendling]

Look around format_decision_state.py:341. That bit of code is trying to split before all named assigns. That bit of code probably isn't triggered for the cases above for some reason (could just be an option isn't set). I think you'll find that either you can modify that bit of code or gain some insight from it on how to address this.

[apbard]

I am not sure we can do that there. Because ideally what we would like to achieve is something like:
"if we have to split somewhere, it is better to split before an argument than in the middle of a list"
In the function _MustSplit we can just force the splitting, right?