BUG: Series.quantile dtype / nan handling issue

[sinhrks]

Found some issues when working on #12572. Both being fixed.

Code Sample, a copy-pastable example if possible

1. DataFrame may coerce to float even if interpolation

# OK, same as numpy
pd.Series([1, 2, 3]).quantile([0.5], interpolation='lower').dtypes
# dtype('int64')

# NG, it must be int
pd.DataFrame({'A': [1, 2, 3]}).quantile([0.5], interpolation='lower').dtypes
# A    float64
# dtype: object

2. May return scalar even if its input is a list-like

# NG, it must be a Series
pd.Series([]).quantile([0.2, 0.3])
# nan

output of pd.show_versions()

Current master.