Switch is_directed_acyclic_graph to use toposort internally

[mtreinish]

Previously the retworkx.is_directed_acyclic_graph() function was
internally using the petgraph::algo::is_cyclic_directed function to look
for a potential cycle in the graph. However, as noted in the docs[1]
this function is recursive and as reported in Qiskit/qiskit#5502
with very large graphs this is causing segfaults. To avoid this
potential issue this commit changes the internal implementation to use
petgraph::algo::toposort which isn't recursive and should be more
reliable.

Fixes: Qiskit/qiskit#5502

[1] https://docs.rs/petgraph/0.5.1/petgraph/algo/fn.is_cyclic_directed.html

[coveralls]

Pull Request Test Coverage Report for Build 415935664

• 3 of 3 (100.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.002%) to 94.851%

---

Totals
Change from base Build 415918432:	0.002%
Covered Lines:	3095
Relevant Lines:	3263

---

💛 - Coveralls

[mtreinish]

I ran some benchmarks on this locally, and using petgraph::algo::toposort ends up being roughly 30% slower for the cases they both worked, but does not segfault for large graphs. While using the petgraph::algo::is_cyclic_directed is segfaulting at >=100000 nodes in a graph. So it's a necessary performance hit and one most people probably won't see because because for a 1000 node graph it still only takes 21μs.