Added BSD-license and improved performance of estimators

The nb_glm estimator now allows to specify whether float32 or float64 should be used internally.
It now also automatically casts input data, i.e. it is no longer necessary to manually change the input data type.