arrow

ARROW-509: [Python] Add support for multithreaded Parquet reads

I'm getting very nice speedups on a Parquet file storing a ~4.5 GB dataset:

```
In [1]: import pyarrow.parquet as pq

In [2]: %time table = pq.read_table('/home/wesm/data/airlines_parquet/4345e5eef217aa1b-c8f16177f35fd983_1150363067_data.0.parq')
CPU times: user 8.21 s, sys: 468 ms, total: 8.68 s
Wall time: 8.68 s

In [3]: %time table = pq.read_table('/home/wesm/data/airlines_parquet/4345e5eef217aa1b-c8f16177f35fd983_1150363067_data.0.parq', nthreads=4)
CPU times: user 8.84 s, sys: 4.28 s, total: 13.1 s
Wall time: 3.91 s

In [4]: %time table = pq.read_table('/home/wesm/data/airlines_parquet/4345e5eef217aa1b-c8f16177f35fd983_1150363067_data.0.parq', nthreads=8)
CPU times: user 13.3 s, sys: 1.15 s, total: 14.4 s
Wall time: 2.86 s
```

This requires a bugfix in parquet-cpp that will come soon in a patch for PARQUET-836

Author: Wes McKinney <wes.mckinney@twosigma.com>

Closes #301 from wesm/ARROW-509 and squashes the following commits:

9816689 [Wes McKinney] Update docs slightly, flake8 warning
239b086 [Wes McKinney] Add support for nthreads option in parquet::arrow, unit tests