Add GC support to graph objects

[mtreinish]

Right now the 2 graph classes in retworkx, PyGraph and PyDiGraph, did
not integrate with the Python garbage collector. This would result in
memory not being cleared in a running python process even if retworkx
objects were only created in temporary context. This commit attempts to
fix this by implementing the garbage collection protocol [1] which
should enable python to clean up retworkx objects when they're no longer
being used.

[1] https://pyo3.rs/v0.12.4/class/protocols.html?highlight=garbage#garbage-collector-integration

[mtreinish]

I'm marking this on hold because I want to do some performance benchmarks on this before we merge it. But this will be a release blocker because it's causing a memory leak in qiskit-terra.

[coveralls]

Pull Request Test Coverage Report for Build 614503023

• 30 of 30 (100.0%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.02%) to 96.465%

---

Totals
Change from base Build 612372672:	0.02%
Covered Lines:	5786
Relevant Lines:	5998

---

💛 - Coveralls

[mtreinish]

Removing the on hold because I think there really isn't any performance impact here at least from the retworkx side. I ran the transpile benchmark in https://github.com/mtreinish/retworkx-bench and there was no real difference in performance.

[itoko]

LGTM as a fix for PyGraph and PyDiGraph, but I wonder if we might have other classes that should implement PyGCProtocol trait. What about WeightedEdgeList and BFSSuccessors? (I've grepped "PyObject" and "pub", and found these two.)
==> Oops, sorry. I've just found you added them in #260

[mtreinish]

LGTM as a fix for PyGraph and PyDiGraph, but I wonder if we might have other classes that should implement PyGCProtocol trait. What about WeightedEdgeList and BFSSuccessors? (I've grepped "PyObject" and "pub", and found these two.)
==> Oops, sorry. I've just found you added them in #260

Yep, I did it for the custom return types separately in #260 because I had been wanting __str__ for those return types for some time, so I handled both at the same time there for all 4 of the custom return types and kept it separate from the graph classes.

[mtreinish]

It looks like there is an infrastructure issue with travis CI on ppc64le nodes (the logs just stop without error) I'm going to merge this