Replace factory functions with classes.

[udifuchs]

Until now Element and ElementTree were factory functions. Element() returned an _Element class and
ElementTree() returned an _ElementTree class.

This PR turns them into classes. These classes should behave exactly the same as the factory function. Specifically, in both cases:

>>> element = Element("test")
>>> type(element)
<class 'lxml.etree._Element'>
>>> callable(Element)
True

The difference is that before this change we could not use isinstance:

>>> isinstance(element, etree.Element)
TypeError: isinstance() arg 2 must be a type, a tuple of types, or a union

While now:

>>> isinstance(element, etree.Element)
True
>>> issubclass(_Element, Element)
True

These checks work because _Element is registered as a virtual subclass of Element.

The motivation for this PR is to support type annotations.

Currently, type stubs for lxml need to annotate the factory functions like this:

def Element(...) -> _Element: ...

Furthermore, the _Element class members are all annotated. But _Element was supposed to be kept as an internal implementation.

With this PR, the Element class can be used as an interface class. This will allow us to create modified stubs for lxml where the _Element class is not mentioned at all.

As far as I can tell, this change should not break backward compatibility. Someone would have to go out of their way to test that Element is a function and not a class. As mentioned above, both are callable.

Still, I must admit that I am not convinced that this is the correct approach. Specifically, it seems a bit of a hack to have an Element class that ends up creating an _Element class.