BUG? Consistency of column parsing

[chris-b1]

xref #11328

In [100]: df = pd.DataFrame({0: range(5), 1:range(5)})

In [101]: df.to_excel('tmp.xlsx')

In [102]: df.to_csv('tmp.csv')

In [103]: pd.read_excel('tmp.xlsx', index_col=0).columns
Out[103]: Int64Index([0, 1], dtype='int64')

In [106]: pd.read_csv('tmp.csv', index_col=0).columns
Out[106]: Index([u'0', u'1'], dtype='object')

[jreback]

I think csv should infer this, though wonder how much code this would break. can you see in the existing suite how much?