Drops duplicate edges in non-MultiGraph PLC SGGraph instances

[rlratzel]

Graph input with duplicate edges intended for Graph/DiGraph instances resulted in internal PLC SGGraph instances with duplicate edges, which were effectively treated as MultiGraphs and caused incorrect results from algorithms like pagerank.

This PR sets the drop_multi_edges PLC SGGraph ctor option to have PLC remove duplicate edges on SGGraph creation.

The overhead to drop duplicate edges for non-MultiGraphs is negligible, and in the case of a large test graph (wikipedia data, 37.5M nodes, 464.5M edges) resulted in an overall speedup for pagerank going from 12.2 seconds to 10.7 seconds on my workstation, likely due to fewer edges to process.

A test was added that uses pagerank to ensure Graphs vs. MultiGraphs are handled correctly and duplicate edges are dropped as needed. The results when run without drop_multi_edges set:

>       assert actual_pr_for_G == approx(expected_pr_for_G)
E       assert {0: 0.0875795...7955580949783} == approx({0: 0....32 ± 1.8e-07})
E
E         comparison failed. Mismatched elements: 4 / 4:
E         Max absolute difference: 0.08785887916592061
E         Max relative difference: 0.5007959662968462
E         Index | Obtained            | Expected
E         0     | 0.08757955580949783 | 0.17543839772251532 ± 1.8e-07
E         1     | 0.41242048144340515 | 0.32456160227748454 ± 3.2e-07
E         2     | 0.41242048144340515 | 0.32456160227748454 ± 3.2e-07
E         3     | 0.08757955580949783 | 0.17543839772251532 ± 1.8e-07

The same test passes when run with the changes in this PR to set drop_multi_edges.

[jnke2016]

Nice. All you did was setting a flag and no performance degradation