This patch also adds support for understanding super calls.
astroid understands the zero-argument form of super, specific to
Python 3, where the interpreter fills itself the arguments of the call. Also, we
are understanding the 2-argument form of super, both for bounded lookups
(super(X, instance)) as well as for unbounded lookups (super(X, Y)),
having as well support for validating that the object-or-type is a subtype
of the first argument. The unbounded form of super (one argument) is not
understood, since it's useless in practice and should be removed from
Python's specification. Closes issue #89.

Inference objects are similar with AST nodes, but they can be obtained
only after inference, so they can't be found inside the AST tree.
Their purpose is to handle at astroid level some operations which can't
be handled when using brain transforms. For instance, the first object
added is FrozenSet, which can be manipulated at astroid's level
(inferred, itered etc). Code such as this 'frozenset((1,2))'
will not return an Instance of frozenset, without having access to its
content, but a new objects.FrozenSet, which can be used just as a nodes.Set.

Enum values should share the same base classes as their defining class.
If this is not the case it may lead to wrong inference results when an
enum member is used - e.g. for the following snippet:

class X(enum.IntEnum):
    one = 1

print([1, 2][X.one])

pylint will detect a "invalid-sequence-index" error as the __index__
method of X.one is not detected.

In Python 3.4+, the module level functions are retrieved with getattr
from a context object, leading to many no-member errors
in Pylint. This patch ensures us that we can retrieve those attributes,
no matter what.