Make `PyObject` parametric?

Thinking about conversion between Julia and Python values and semantics, I wonder if we could make PyObject parametric, so make it PyObject{P <: PyConversionPolicy} where P encodes how we want to interpret the object.

For example, PyObject{Nice} would be equivalent to PyObject as it is now, with user-friendly automatic conversions. PyObject{Raw} would perform no automatic conversion at all, which would be useful for wrapping Python packages into Julia, where complete control is required. Other possibilities could be PyObject{Numpy} which converts Julia arrays to numpy arrays instead of to lists, for example.

This policy parameter would be inherited in a natural way, so operations on objects with the same policy produce an object with that policy.

I don't know if the effort to implement this would be worth it, since you can already achieve the same things using PyCall's clunkier syntax, but I just wanted to gather opinions and suggestions.

Empieza leyendo el comportamiento de conversión existente de PyObject y la sintaxis de PyCall con la que se compara el issue. Aclara si se desea una PyConversionPolicy con las variantes Nice, Raw y Numpy, incluida la herencia de la policy entre operaciones; el trabajo estaría terminado cuando haya acuerdo sobre el diseño y su semántica de conversión.