Pandas Nusicance Columns in cumulative aggregations

We're inconsistent with excluding nuisance columns

In [37]: df = pd.DataFrame({
    ...:     "A": [1, 2, 3, 4],
    ...:     "B": pd.interval_range(0, 4),
    ...: })

In [38]: df.dtypes
Out[38]:
A     int64
B    object
dtype: object

In [39]: df.sum()
Out[39]:
A    10
dtype: int64

In [40]: df.cumsum()
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-40-1c578612e53e> in <module>()
----> 1 df.cumsum()

~/Envs/pandas-dev/lib/python3.6/site-packages/pandas/pandas/core/generic.py in cum_func(self, axis, skipna, *args, **kwargs)
   7781             mask = isna(self)
   7782             np.putmask(y, mask, mask_a)
-> 7783             result = accum_func(y, axis)
   7784             np.putmask(result, mask, mask_b)
   7785         else:

~/Envs/pandas-dev/lib/python3.6/site-packages/pandas/pandas/core/generic.py in <lambda>(y, axis)
   7400         cls.cumsum = _make_cum_function(
   7401             cls, 'cumsum', name, name2, axis_descr, "cumulative sum",
-> 7402             lambda y, axis: y.cumsum(axis), "sum", 0., np.nan)
   7403         cls.cumprod = _make_cum_function(
   7404             cls, 'cumprod', name, name2, axis_descr, "cumulative product",

TypeError: unsupported operand type(s) for +: 'pandas._libs.interval.Interval' and 'pandas._libs.interval.Interval'

I'll update this with more dtypes (datetime, period, categorical) and more aggregations. Ideally we'd be consistent between the cumulative and non-cumulative versions.

Comment From: tdpetrou

Also very closely related is accumulation with missing values.

>>> df = pd.DataFrame({'a': [1, np.nan, 5]})
>>> df.cumsum()
     a
0  1.0
1  NaN
2  6.0

Couldn't this be:

>>> df.cumsum()
     a
0  1.0
1  1.0
2  6.0