Overhaul AbstractGraphIndices interface.

[jack-dunham]

Most changes pertain to the files quotientvertex.jl and quotientedge.jl.

1. Base.getindex now canonize the index argument via to_graph_index before calling get_graph_indices or get_graph_index when canonized to a subtype of AbstractGraphIndices or not, respectively.
2. The function get_graph_index throws a MethodError by default, therefore using the Base.getindex syntax with a single vertex as the index, e.g. g[1], now throws a MethodError for AbstractNamedGraph graphs.
   • One should use g[Vertices([1])] or subgraph(g, 1) to construct a subgraph of g with only the vertex 1.
   • This is so Base.getindex behaves similarly for AbstractNamedGraph and AbstractDataGraph as AbstractDataGraph specializes on get_graph_index.
3. One can use Dictionaries.getindices to directly call get_graph_indices no matter the argument. The index argument then gets canonized via to_graph_indices.
4. All QuotientXX index collection structs now have an edge counterpart.

[mtfishman]

@jack-dunham looks like this requires a version bump. I'm torn about whether we should consider this to be breaking, maybe to be on the safe side we should mark it as breaking.

[jack-dunham]

@jack-dunham looks like this requires a version bump. I'm torn about whether we should consider this to be breaking, maybe to be on the safe side we should mark it as breaking.

I think breaking is sensible as indexing like g[1] and g[[1,2]] no longer behaves as it did previously. The former should error and the latter assumes [1,2] is a vertex, and thus should also error. Both of these constructed subgraphs previously.