Custom ingredient types

• It is now possible to extend PyPlant with custom ingredient types.
  • Subclass IngredientTypeSpec, implement abstract methods and register with plant.warehouse.register_ingredient_specs.
• HDF-arrays, Keras models and SciPy sparse matrices are now ingredient specs, available in pyplant.specs

Ingredient type inference.

• No longer have to provide an ingredient type, when sending it.
• Pickled NumPy arrays are now allowed.
• More robust against data corruption.
• Improved unit test coverage.

First public version

Published on PyPI.
Wrote a proper introduction guide.
Cleaned up some names.