Skompiler

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

- Changed skompile call signature
- Fixed Excel code generation
- Changed SQLAlchemy translator
- multistage=True is the default now for SQL and Excel
- Improved code organization
- Nicer DSL for generating SKAST, basic symbolic processing and type checking
- New algorithms: AdaBoostClassifier, KMeans, PCA, MLPClassifier, MLPRegressor,
Normalizer, StandardScaler, MinMaxScaler, MaxAbsScaler, Binarizer
- (0.5.1) Rudimentary support for Keras MLP models.
- (0.5.2) Fixes issue 1
- (0.5.3) Fixes for Python 3.5, removed dependency on staticdispatch.
- (0.5.4) Portable Format for Analytics (PFA) added as a target
- (0.5.5) Translation to SQLAlchemy supports CTE as from_obj

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

- Multi-stage Excel code generation

-----------

- Fixed warnings from SKlearn 1.0 (PR11, PR14).
- The internal TreeWalker class now tolerates excess input variables. Also from PR14.

-----------

- Removed Keras translation (to avoid dependency on Tensorflow)
- Modernized dependencies. SKLearn >= 0.22 required as of this moment.

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

- More logical set of SK AST nodes, proper handling of Let nodes
- Support for Pipeline objects

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

- Multi-stage SQL code generation