Turbodbc

-------------

* Release GIL while fetching batches using Apache Arrow engine
* Support Arrow 0.14.X

-------------

* Correctly report odbc errors when freeing the statement handle as exceptions;
see `Github issue 153 <https://github.com/blue-yonder/turbodbc/issues/153>`_
(thanks byjott)
* Support user-provided gmock/gtest, e.g. in conda environments via
``conda install -c conda-forge gtest gmock``.
* Make source code compatible with Apache Arrow 0.13.0

-------------

* Update to Apache Arrow 0.12
* Support the unicode datatype in the Arrow support. This primarily enables
MS SQL support for the Arrow adapter.
* Windows support for the Arrow adapter.
* Add a new entry to the build matrix that tests Python 3.7 with conda and
MS SQL on Linux.
* Big hands to xhochy for making all these changes!

-------------

* Adjust generators to conform to PEP-479
* Build wheels for Python 3.7 on Windows
* Drop support for Python 3.4
* Update to Apache Arrow 0.11

-------------

* Added new keyword argument ``fetch_wchar_as_char`` to ``make_options()``.
If set to ``True``, wide character types (``NVARCHAR``) are fetched and
decoded as narrow character types for compatibility with certain
databases/drivers (thanks yaxxie).
* Added batched fetch support for Arrow as ``fetcharrowbatches()``
(thanks mariusvniekerk).
* Support (u)int8, (u)int16, (u)int32 Arrow columns on
``executemanycolumns()`` (thanks xhochy).

-------------

* Added support for ``with`` blocks for ``Cursor`` and ``Connection``
objects. This makes turbodbc conform with
`PEP 343 <https://www.python.org/dev/peps/pep-0343/>`_
(thanks AtomBaf)
* Added new keyword argument ``force_extra_capacity_for_unicode`` to
``make_options()``. If set to ``True``, memory allocation is modified
to operate under the assumption that the database driver reports field
lengths in characters, rather than code units (thanks yaxxie).
* Updated Apache Arrow support to work with both versions 0.8.0 and 0.9.0
(thanks pacman82)
* Fixed a bug that led to ``handle limit exceeded`` error messages when
``Cursor`` objects were not closed *manually*. With this fix, cursors
are garbage collected as expected.