Datarobot-batch-scoring

====================
* use chardet module `chardet <https://pypi.python.org/pypi/chardet>`_ to
attempt to detect character encoding

* use standard lib csv module to attempt to discover CSV dialect

* use stream decoder and encoder in python 2 to transparently convert to utf-8

* provide a mode for sending all user messages to stdout

================
* separate process for disk IO and request payload serialization

* avoid codecs.getreader due to IO bottleneck

* dont parse CSV (fail fatally on multiline csv)

* multiline mode (to be renamed)

* keep_cols resolution

===========================

* Get rid of gevent/asyncio, use thread-based networking

* Show path to logs on every unexpected error

* Convert cmdline argument parser from docopt to argparse

* Add configuration file support

* Refactor logging/ui

* Drop support of making predictions using 'v2' Modeling API

=====

* Fix bug under Python 2 where gevent was fatally failing on timeouts.

* Added timeout argument.

* Both asyncio and gevent now retry within the request exception handler.

* Authorization now checks schema too and thus we fail much earlier if
input not correct.

=====

* Fix bug under Python 2 where gevent was silently dropping batches.

* Better checks if run completed successfully.

* Fail fast on missing column or dtype mismatch.

* Add naming of prediction column for regression.

* Fix ignore datarobot_key.

=====

* Update requirements for Python 3 to minimum versions.