Bug in internals alignment for Series.combine_first with Extension dtype

[TomAugspurger]

In [5]: a = pd.Series(pd.Categorical([0, 1, 2], categories=list(range(5))))

In [6]: b = pd.Series(pd.Categorical([2, 3, 4], categories=list(range(5))), index=[2, 3, 4])

In [7]: a.combine_first(b)
Out[7]:
0    0.0
1    1.0
2    2.0
3    NaN
4    NaN
dtype: category
Categories (5, int64): [0, 1, 2, 3, 4]

Compare with the expected (aside from dtype)

In [8]: a = pd.Series([0, 1, 2])

In [9]: b = pd.Series([2, 3, 4], index=[2, 3, 4])

In [10]: a.combine_first(b)
Out[10]:
0    0.0
1    1.0
2    2.0
3    3.0
4    4.0
dtype: float64

Something is going wrong inside Block.apply at https://github.com/pandas-dev/pandas/blob/master/pandas/core/internals/managers.py#L386-L387

(Pdb) pp b.mgr_locs.indexer
slice(0, 1, 1)
(Pdb) pp self.items[b.mgr_locs.indexer]
Int64Index([0], dtype='int64')

that should be

(Pdb) pp b.mgr_locs.indexer
slice(0, 5, 1)
(Pdb) pp b_items
Int64Index([0, 1, 2, 3, 4], dtype='int64')

I'm hitting this in the DatetimeArray refactor.

I suspect that this is a symptom of #23023