Conflicts:
	doc/source/reference/api_2.0.rst
	doc/source/reference/credits.rst

Also, use the Numpy that is in the Pandas namespace. Will be the same thing,
but saves an import.

The from_pandas_dataframe function was modified to use a generator that
yields an ebunch for every entry in the DataFrame. In this way, the DataFrame
does not need to be square.

Adds the `parallel_edges` keyword argument to the `from_numpy_matrix`
and `from_scipy_sparse_matrix`. This allows the user to specify whether
to interpret an integer matrix as parallel edges in a multigraph or
single weighted edges in a multigraph.

This restores the default behavior of these functions that was changed
by pull request #1305.

Previously in `from_scipy_sparse_matrix()` and `from_numpy_matrix()`, if
`create_using` were an undirected multigraph and the input matrix were
symmetric, the edge (u, v) and the edge (v, u) would be added
separately, thereby resulting in two parallel edges. This change makes
the function behave more intelligently: it now assumes that when the
input matrix is symmetric and `create_using` is undirected, the edge (u,
v) should be added only once.

Forces from_numpy_matrix() and from_scipy_sparse_matrix() to create a
multigraph with parallel edges if the input adjacency matrix has integer
entries and the `create_using` keyword argument is an instance of
`MultiGraph`. In this case, each positive integer entry in the adjacency
matrix is interpreted as the number of parallel edges, each with weight
1, joining two vertices. Otherwise, if any of these conditions is not
met, the function behaves as it previously did: an integer entry is
interpreted as the weight of a single edge joining the two vertices.

This is not a backwards compatible change: it changes the behavior of
the function if the input is an integer matrix and `create_using` is a
multigraph.