Now default constructed edge descriptors are impossible to occur in
regular graphs.

This also clarifies the relevant documentation.

This switches the edge_descriptor type in adj_list<> from std::tuple to
a custom struct.

This also implements direct manipulation of OpenMP settings (schedule
and number of threads) via python.

Now, if either vertex or edge filtering is enabled, simultaneous
filtering is activated. By not considering vertex or edge filtering in
completely independent fashion, the compile time memory requirements
reduced to about 50%.