arrow

ARROW-3903: [Python] Random array generator for Arrow conversion and Parquet testing

Generate random schemas, arrays, chunked_arrays, columns, record_batches and tables.
Slow, but makes quiet easy to isolate corner cases (already created jira issues). In follow up PRs We should use these strategies to increase the coverage. It'll enable us to reduce the issues, We could even use it for generate benchmark datasets periodically (only if We persist somewhere).

Example usage:

Run 10 samples (dev profile):
`pytest -sv pyarrow/tests/test_strategies.py::test_tables --enable-hypothesis --hypothesis-show-statistics --hypothesis-profile=dev`

Print the generated examples (debug):
`pytest -sv pyarrow/tests/test_strategies.py::test_schemas --enable-hypothesis --hypothesis-show-statistics --hypothesis-profile=debug`

Author: Krisztián Szűcs <szucs.krisztian@gmail.com>

Closes #3301 from kszucs/ARROW-3903 and squashes the following commits:

ff6654cc <Krisztián Szűcs> finalize
8b5e7eae <Krisztián Szűcs> rat
61fe01d2 <Krisztián Szűcs> strategies for chunked_arrays, columns, record batches; test the strategies themselves
bdb63df6 <Krisztián Szűcs> hypothesis array strategy