Make _ext_getattr handle a default, and add _ext_replacement_getattr …

[jamadden]

…as the unoptimized version. See #73

[cutz]

This looks good. Should we add a _ext_replacement_setattr for parity?

[jamadden]

This package doesn't actually ever pass that argument, so even if subclasses were already overriding that method, there shouldn't be any compatibility issues.

[jamadden]

Should we add a _ext_replacement_setattr for parity?

My instinct tells me that while one-off uses (outside of a loop, where the optimization of caching the return value of _ext_replacement doesn't matter) of _ext_getattr (once getattr is replaced) may be relatively common in subclasses, one-off uses of _ext_setattr shouldn't be...setting multiple fields should happen through the interface definition and thus already be handled by these superclasses (or be a standard field, which again is already handled).