[Regression] yapf version 0.20.0 Facebook style is broken in yapf verison 0.29.0

Related to #579

There's a YAPF demo being used officially in the README.md: https://yapf.now.sh/

Source code of the online demo: https://github.com/jpadilla/yapf-online

If we take a look at the Pipfile: https://github.com/jpadilla/yapf-online/blob/master/Pipfile

We can see the online demo is using yapf version 0.20.0.

This is what the "facebook" style looks like in 0.20.0:


Notice that in the yapf-formatted code (below), the array list of dictionary entries is neatly split, and there is no trailing comma in the list.

```
x = {
'a': [
{
'a': {
'a': 37,
'b': 42,
'c': 927
},
'b': 42,
'c': 927
}, {
'a': 37,
'b': 42,
'c': 927
}, {
'a': 37,
'b': 42,
'c': 927
}
]
}

y = 'hello ' 'world'
z = 'hello ' + 'world'
a = 'hello {}'.format('world')

class foo(object):
def f(self):
return 37 * -2

def g(self, x, y=42):
return y

def f(a):
return 37 - a[42 - x:y**3]

```

I'm currently using yapf version 0.29.0, installed using Python 3.8.0 64-bit, through `python -m pip install yapf`. The "facebook" style used in yapf 0.29.0 is incorrectly formatted, as shown in the code below. The code is identical to the code used in the online demo:

```
x = {
'a':
[
{
'a': {
'a': 37,
'b': 42,
'c': 927
},
'b': 42,
'c': 927
},
{
'a': 37,
'b': 42,
'c': 927
},
{
'a': 37,
'b': 42,
'c': 927
},
]
}

y = 'hello ' 'world'
z = 'hello ' + 'world'
a = 'hello {}'.format('world')

class foo(object):
def f(self):
return 37 * -2

def g(self, x, y=42):
return y

def f(a):
return 37 - a[42 - x:y**3]

```

Notice how the array list of dictionary entries are formatted in a weird way, and I think this is a regression from what is essentially a neatly formatted code.

No yapf_test test cases were failing, and PEP8 did not care about the indentation layout of the array list. However, from a readability standpoint, near the closing bracket of the array list, you will see an incorrectly indented ], which is indented twice.

That is the main problem of this issue. There seems to be some inconsistent list formatting that, somehow, the test cases aren't catching.