BUG: Correctly specified broadcasting leads to a RecursionError

[rs2]

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of pandas.
• I have confirmed this bug exists on the master branch of pandas.

Reproducible Example

import pandas as pd                                                                                                                          
a = pd.DataFrame(data=[[21] * 3, [22] * 3], columns=pd.MultiIndex.from_tuples([(1, 10), (2, 10), (3, 11)], names=['foo', 'bar']))            
b = pd.DataFrame(data=[[31] * 2, [32] * 2], columns=pd.Int64Index([10, 11], name='bar'))                                                     
print(a)                                                                                                                                            
print(b)
print(b + a)  # This works, but returns an incorrect result: all NaNs
print(a + b)  # This raises a RecursionError

Issue Description

    755             self, other, op, axis, default_axis, fill_value, level
    756         ):
--> 757             return _frame_arith_method_with_reindex(self, other, op)
    758 
    759         other = _align_method_FRAME(self, other, axis)

RecursionError: maximum recursion depth exceeded while calling a Python object

Expected Behavior

In [6]: b + a
Out[6]:
foo   1   2   3
bar  10  10  11
0    52  52  52
1    54  54  54

In [7]: a + b
Out[7]:
foo   1   2   3
bar  10  10  11
0    52  52  52
1    54  54  54

Installed Versions

pandas : 1.0.5

[rs2]

Closed by #33765

[rs2]

Also see #25760