ENH: Add support for hiding edges for callable capacity or weight parameter

Current Behavior

G = nx.DiGraph()
G.add_edge(0, 1, weight=1)
G.add_edge(1, 2, weight=2)
def myweight(u, v, e_attr):
    if u == 0 and v == 1:
        return e_attr['weight']
    else:
        return None
nx.bellman_ford_predecessor_and_distance(G, source=0, weight=myweight)

throws TypeError: unsupported operand type(s) for +: 'int' and 'NoneType'

Expected Behavior

In most of the algorithms like dijkstra, floyd_warshall, if the capacity/weight function returns None, then that edge is assumed to be hidden. It is not currently supported in bellman_ford. We also need to find if there are other such functions, where the parameter can be weight/capacity function but hiding an edge is not supported.

Additional context

Comments in #8440 https://github.com/networkx/networkx/pull/8440#issuecomment-3721614854