Reuse PyEnumerable, removing need for PyKeyValuePairEnumerable

[atifaziz]

PyKeyValuePairEnumerable<,> had almost an identical implementation to PyEnumerable<> except for the conversion of the PyObject value. The former calls PyObject.As<,>:

CSnakes/src/CSnakes.Runtime/Python/PyKeyValuePairEnumerable.cs

Lines 42 to 43 in b9a6b8d

	using PyObject pyObject = PyObject.Create(result);
	current = pyObject.As<TKey, TValue>();

whereas the latter calls PyObject.As<>:

CSnakes/src/CSnakes.Runtime/Python/PyEnumerable.cs

Lines 42 to 43 in b9a6b8d

	using PyObject pyObject = PyObject.Create(result);
	current = pyObject.As<TValue>();

That's a lot of duplication to maintain, sync, test and cover. This PR consolidates the duplication so there's only one bulk implementation. It does this by adding another type parameter (called TImporter) to PyEnumerable<> that determines how the PyObject is imported (into presumably a managed type):

internal class PyEnumerable<TValue, TImporter> : IEnumerable<TValue>, IEnumerator<TValue>, IDisposable
    where TImporter : IPyObjectImporter<TValue>
{
    // omitted for brevity
}

The importer has a static member, which bears zero cost (that is, no allocation and no virtual dispatch needed):

interface IPyObjectImporter<out T>
{
    static abstract T Import(PyObject pyObj);
}

The two import strategies, PyObject.As<> and PyObject.As<,>, are then embodied into implementations of this interface. The import strategy is chosen at the time PyEnumerable<,> is instantiated. A PyEnumerable<> with a single type parameter still exists, but which uses the default strategy of PyObject.As<> so most of the code is untouched except in PyDictionary<,>.GetEnumerator.

[tonybaloney]

Love it! this is a lot cleaner