Wrong dtype when mixed dtype DataFrame is accessed with complete indexer

[toobaz]

A small, complete example of the issue

In [2]: df = pd.DataFrame({'float' : range(4),
   ...:                    'int'   : range(4)},
   ...:                   index=pd.MultiIndex.from_product([[1,2]]*2))

In [3]: df['float'] = df['float'].astype(float)

In [4]: df.loc[(1,), 'int']
Out[4]: 
1    0
2    1
Name: int, dtype: int64

Right!

In [5]: df.loc[(1,), 'int'].loc[1]
Out[5]: 0

Right!

In [6]: df.loc[(1,1), 'int']
Out[6]: 0.0

Wrong!

Expected Output

Still an integer.

Output of pd.show_versions()

In [7]: pd.show_versions() ## INSTALLED VERSIONS

commit: None
python: 3.5.1.final.0
python-bits: 64
OS: Linux
OS-release: 4.5.0-2-amd64
machine: x86_64
processor:
byteorder: little
LC_ALL: None
LANG: it_IT.utf8
LOCALE: it_IT.UTF-8

pandas: 0.19.0+1.g96b364a
nose: 1.3.7
pip: 1.5.6
setuptools: 18.4
Cython: 0.23.4
numpy: 1.10.4
scipy: 0.16.0
statsmodels: 0.8.0.dev0+f80669e
xarray: None
IPython: 5.1.0.dev
sphinx: 1.4.5
patsy: 0.3.0-dev
dateutil: 2.2
pytz: 2012c
blosc: None
bottleneck: 1.1.0
tables: 3.2.2
numexpr: 2.5
matplotlib: 1.5.1
openpyxl: None
xlrd: 0.9.4
xlwt: 1.1.2
xlsxwriter: 0.7.3
lxml: None
bs4: 4.4.0
html5lib: 0.999
httplib2: 0.9.1
apiclient: 1.5.0
sqlalchemy: 1.0.11
pymysql: None
psycopg2: None
jinja2: 2.8
boto: 2.38.0
pandas_datareader: 0.2.1

[toobaz]

Oh, my example above is actually unnecessarily complicated - the index can be a normal Int64Index, as in

In [8]: df = pd.DataFrame({'float' : range(4),
   ...:                    'int'   : range(4)})

In [9]: df['float'] = df['float'].astype(float)

In [10]: df.loc[2, 'int']
Out[10]: 2.0

[jreback]

duplicate of #11617