Make isomorphism algorithm generic and add subgraph isomorphism check

[althonos]

Hi there!

I refactored the code in isomorphism.rs to make the algorithms generic, and I also added functions to check for subgraph isomorphism as requested in #48.

Added

• petgraph::visit::EdgeCount trait to get the edge count for a graph (which is needed to check for the edge count before actually the VF2 matching algorithm), which is implemented for all graph implementations.
• petgraph::algo::is_isomorphic_subgraph that checks if a graph is isomorphic to a subgraph of another graph (and petgraph::algo::is_isomorphic_subgraph_matching for the semantic matching variant)

Changed

• petgraph::algo::is_isomorphic is now generic over its two input graphs, meaning that they don't even have to be of the same concrete type. There is however a check at compile time to make sure both graphs are either Directed or Undirected.
• Semantic checking can be performed even if both graphs don't have the same data type, as long as the node_match and edge_match callbacks are properly typed.
• The private code of is_isomorphic is now properly encapsulated, whereas the old code was doing everything in one big function with inline closures.
• I inlined some implementations of NodeIndexable trait as not doing so had an impact on performance (the generic code uses to_index and from_index methods much more)

Benchmarks

graph.bench are benchmarks with the code from the master branch, generic.bench with the code from this PR.

>>> cargo benchcmp graph.bench generic.bench 
 name                       graph.bench ns/iter  generic.bench ns/iter  diff ns/iter  diff %  speedup 
 full_iso_bench             1,582                1,549                           -33  -2.09%   x 1.02 
 petersen_iso_bench         1,277                1,299                            22   1.72%   x 0.98 
 petersen_undir_iso_bench   913                  898                             -15  -1.64%   x 1.02 
 praust_dir_no_iso_bench    1,102,471            1,120,120                    17,649   1.60%   x 0.98 
 praust_undir_no_iso_bench  1,052,307            1,047,779                    -4,528  -0.43%   x 1.00

To do

• Figure out a way for is_isomorphic not to require the input to be DataMap (probably by changing the SemanticMatcher trait).
• Figure out a way to perform semantic feasibility validation for edges in a generic manner: currently the code needs to access an edge given its two endpoints, (this is causing the iso_matching test to fail currently)
• Figure out a way for is_isomorphic not to require IntoEdgesDirected for the semantic validation of edges (IntoNeighborsDirected only is required for the rest of the algorithm).
• Add a benchmark to assess the performance of semantic-checking isomorphism functions.

[XVilka]

Wow, amazing. On the quick look it is good, but I need more time to read it thoroughly and test. Good job!

[althonos]

@XVilka ping me if you need me to rebase / change anything 😉

[XVilka]

Please next time don't include whitespace changes - it makes review process harder, especially with those huge graph maps of 1 and 0.

[althonos]

Please next time don't include whitespace changes - it makes review process harder, especially with those huge graph maps of 1 and 0.

Sorry about that, my editor removes trailing whitespaces and I always notice it too late.