Detect "dud" qubits from the noise model and exclude them from the architecture before compiling

[cqc-alec]

Seems it can happen that we get a noise model from IBMQ that includes "dud" qubits, which have readout errors of [[0.0, 1.0], [1.0, 0.0]]. Can we detect these and make sure we don't compile onto them?

See #449 .