============
- use SQLAlchemy and support Postgres (other engines can easily be added)
- specifying variable environment is optional, by default it assumes `locals()` and `globals()` together
- new class-based interface: allows not to specify database each time
- cleaner interface: docstring, custom exception, refactoring
- fix datetime handling
- don't drop column named 'index'
- drop implicit automatic conversion to dataframes: it is easy for user to call `pd.DataFrame` constructor, so what for the magic?
- testing uses pytest and tests for all supported backends
- add asserts for actual content in tests, not length only
- demo fixed to work with modern pandas and python versions
- add dependencies to `setup.py`