Alosi

* Add twine and setuptools_scm packages to docker-based build environment

* refactor data-related utilities (olx generation, course data loading via api) to optional data subpackage
* olx generation with jinja template based component
* fix inverse odds utility function in engine module

Major:
* `learner_mastery_odds` in `get_recommend_params()` output replaced with log odds mastery, `L`
* `recommendation_score()` takes guess and slip as input instead of relevance
* recommendation subscores (P, R, D) take `L` as input instead of mastery odds

Patch:
* Use setuptools_scm for versioning in `setup.py`
* Fill `NaN` in `calculate_relevance()`
* Fill `NaN` for difficulty in `recommendation_score()`

See https://github.com/harvard-vpal/alosi/commit/6ab96168bb281f52bcf5bf282b64ef035d98dd7e

Fix bug in api calls in `engine_api` module where extra url arguments were passed to requests call.

* Text encoding for google sheets table data (https://github.com/harvard-vpal/alosi/pull/4)
* OLX course export builder (https://github.com/harvard-vpal/alosi/pull/5)
* Add base api client (https://github.com/harvard-vpal/alosi/pull/6)