re-implement Safra's construction
Right now we have an implementation of Safra's construction in tgba/tgbasafracomplement.hh. However this implementation uses the old, abstract interface, which is slow and pointless (we are never going to complement something built on-the-fly). So this should be reimplemented using the new interface: both reading and producing an explicit graph.
We might want to wait until we have a way to store Rabin or Streett automata, and have distinct steps for T(G?)BA->DRA and DRA->BA.
There is also some work on generalizing Safra's construction to TGBA that might be worth implementing. Reading this thesis, it's not clear to me if the generalization of Safra is any better than degeneralizing to TBA before running the complement.