DOC/TST: update the parquet (pyarrow >= 0.15) docs and tests regarding Categorical support

[jorisvandenbossche]

Wes is doing great work in Apache Arrow on parquet's categorical support, which means that roundtripping to parquet with to_parquet/read_parquet will preserve categorical dtypes (and with a much better performance as before).

See https://issues.apache.org/jira/browse/ARROW-3246 (and linked issues), apache/arrow#5110

We will need to:

• update the tests for pyarrow to test this faithful roundtrip (depending on the pyarrow version):

  pandas/pandas/tests/io/test_parquet.py

  Line 409 in 802f670

  # additional supported types for pyarrow

  ,

  pandas/pandas/tests/io/test_parquet.py

  Line 451 in 802f670

  def test_categorical(self, pa):

• update the documentation. Eg the caveats section at https://dev.pandas.io/user_guide/io.html#parquet

[galuhsahid]

I'd like to work on this