SelectSwapQROM maintenance

[mpharrigan]

• autodoc
• fix docstrings to be valid markdown
• remove the use of cirq.inverse which produces cirq._InverseCompositeGate which hurts my head
• while I was at it: I changed the control flow of the construction to be more verbose.
• sneak in a fix to the adjoint/t_complexity shim. Note: something like an Adjoint bloq is incompatible with the (vanilla) t complexity protocol because it involves leaping more than one level in the call graph.

This doesn't change the actual construction, cc #699

[mpharrigan]

Supporting cirq simulation via the bloq decomposition is borked. The error message fails trying to use cirq to simulate an Allocate bloq; but playing around with this I also got errors about a CNOT having the same source and target qubit.