arrow

ARROW-3910: [Python] Set date_as_objects=True as default in to_pandas methods

This does not add a deprecation warning primarily because it's a bit difficult to do (we would need to check the data types whether it's a date -- or in the case of a table, if any field is a date--, and then warn if so). `True` is the correct option though in order to accurately roundtrip data to and from pandas. Some users might have some workarounds floating around, but this is sufficiently advanced stuff already.

With this patch, date data round trips with no special options

```
In [2]: import pyarrow as pa

In [3]: import datetime

In [4]: arr = pa.array([datetime.date(2000, 1, 1), None])

In [5]: arr
Out[5]:
<pyarrow.lib.Date32Array object at 0x0000022CCDB1BBD8>
[
  10957,
  null
]

In [6]: arr.to_pandas()
Out[6]: array([datetime.date(2000, 1, 1), None], dtype=object)

In [7]: pa.array(arr.to_pandas())
Out[7]:
<pyarrow.lib.Date32Array object at 0x0000022CCDC7FE58>
[
  10957,
  null
]
```

If others strongly feel it's worth going to the effort of raising a deprecation warning, please chime in.

Author: Wes McKinney <wesm+git@apache.org>

Closes #3272 from wesm/ARROW-3910 and squashes the following commits:

308afe56 <Wes McKinney> Add Windows makefile for Sphinx, add section about date conversions to pandas.rst
f77c2967 <Wes McKinney> Set date_as_objects=True as default in to_pandas methods