Initial commit for pyarrow support

[sidneymau]

Problem & Solution Description (including issue #)

This PR adds support for pyarrow for

1. IO with parquet files (pyarrow.parquet and pyarrow.dataset)
2. In-memory tabular representation (pyarrow.Table)

There is a lot of overlap between this and pandas (as pandas uses pyarrow as a backend), but in principle pyarrow offers much better scaling for, e.g., reading/writing data in batches, streamed computations (pyarrow.acero), etc., so it seems worth having.

Code Quality

• [ ] My code follows the code style of this project
• [ ] I have written unit tests or justified all instances of #pragma: no cover; in the case of a bugfix, a new test that breaks as a result of the bug has been added
• [ ] My code contains relevant comments and necessary documentation for future maintainers; the change is reflected in applicable demos/tutorials (with output cleared!) and added/updated docstrings use the NumPy docstring format
• [ ] Any breaking changes, described above, are accompanied by backwards compatibility and deprecation warnings

[sidneymau]

A few notes:

• I haven't succeeded in modifying the test suite to cover all of the new pyarrow bits. The conversion from an Astropy Table to a pyarrow Table (which should raise a NotImplementedError) never runs, for example
• Pyarrow/parquet schema supports metadata (both at the schema-level and at the individual field level); I need to add code to propagate metadata to/from the schema appropriately
• I've set the code to use pyarrow when the file format is .parquet and pandas when .pq/.parq. Not sure the best way to resolve this... (I also would opine that it would be easier to only support .parquet as that tends to be the preferred extension...)

[sidneymau]

Linking this PR to #66