Speed improvments

[ebertmi]

Currently, we have some bottlenecks in the implementation:

• PyArray_FromAny seems to use significant time in the array creation process
• MatrixProduct2 uses time in calling Sk.builtin.func.tp$call in dot_f
• PyArray_NewFromDescr_int

Maybe we can just set the prototype of our object and do not need to create a a new instance and pass in all the values as python objects that get remapped to js anyways.