Topological sensitivity in the diagonal compiler

[ecpeterson]

The diagonal compiler included as part of #498 makes its choice of where to insert CNOTs based solely on most frequently sighted ZZ Pauli pairs. This has the benefit of producing low-CNOT-count circuits, but there's no reason to think that they're low-depth or that they do a good job of respecting chip topology.

Investigate this insertion strategy. Can it be modified to be topology-sensitive? Can it be modified to improve circuit depth rather than total CNOT count?

[ecpeterson]

Here's a related paper that also makes use of certain Hamiltonians' large isotropy groups: https://arxiv.org/abs/2104.00728 .