DOC/BUG: broken example in read_parquet with selecting columns

[jorisvandenbossche]

In the dev docs the example that subsets the columns to read with read_parquet is broken for the pyarrow engine: http://pandas-docs.github.io/pandas-docs-travis/io.html#io-parquet

In [514]: result = pd.read_parquet('example_pa.parquet', engine='pyarrow', columns=['a', 'b'])
...
IndexError: Table column index 6 is out of range

In [515]: result = pd.read_parquet('example_fp.parquet', engine='fastparquet', columns=['a', 'b'])

In [516]: result.dtypes
Out[516]: 
a    object
b     int64
dtype: object

This is due to a bug in pyarrow (which I am reporting over there, due to how pyarrow deals with the pandas metadata if not all columns are present), but in the meantime we should also fix our docs to not show this buggy example.

[jorisvandenbossche]

Issue is here: https://issues.apache.org/jira/browse/ARROW-1883 and PR here: apache/arrow#1386

[jorisvandenbossche]

And we maybe should also add a test for this case

[jorisvandenbossche]

The docs are updated to not include this, so removing the 0.21.1 milestone, we let's keep this open to make sure to revert the PR once pyarrow 0.8.0 is released.