Allow decoders to decode Python types derived from primitives

Environment

• Pythonnet version: 3.0.2
• Python version: not relevant for this issue
• Operating System: not relevant for this issue
• .NET Runtime: not relevant for this issue

Details

• We have used the preview version in the past and wrapped it so that we are able to update the package and change against interfaces for future releases. One part of it is converting numpy types. By default such types are not supported anymore in the release version and it seems to be recommended to use codecs instead. I registered a codec (same as pandanet) but PyObject.As<double>() does not convert PyType = <class 'numpy.int32'> with the help of the codec.
  By debugging it never hits the PyObjectConversions.TryDecode because double it is not equal to "object type" and not one of the "DecodableByUserTypes".

  TODO

to reproduce:

    import numpy as np
    numpyArray = np.array([1, 2])

// C# part
var value = numpyArray.GetItem(i)
var valueAsT = value.As<double>(); // fails

Whish:

• solution to handle such a case i.e. by configuring hooks or use the existing codec approach to convert before it hits the usual ToPrimitive(...) function which fails.

Start at PyObject.As() and trace the path through PyObjectConversions.TryDecode, registered codecs, and ToPrimitive(...), using the numpy.int32 reproduction as the first check. Determine where derived Python primitive types should be accepted and define completion as successful conversion of the shown array values to double without breaking existing codec behavior.