API for advanced noise modeling should provide tools for Device-level noise model generation

When writing complex algorithms and proposals a typical workflow looks like this:

1. Write up and debug the code on the noiseless (perfect) QVM
2. Add noise to the circuits and re-evaluate the performance under noisy conditions on the QVM
3. Run it on the QPU

The current API to add advanced noise models to a program do not foster a single transition from step 1 to 2. The reason is as follows:

The QVMConnection does either take in simple X, Y, Z errors after 1Q operations and simple measurement noise, or you have to write a full Device specification which currently is non-trivial as there is no supporting functionality. Moreover, it does not allow to specify noise on gates that are not native to the underlying QPU gateset, e.g. I cannot specify the fidelity for a CNOT gate.

The more usable API to add advanced noise into my programs interleaves with the construction of a pyquil Program and hence forces the developer to re-write large chunks of the code developed in step 1, marking a disruption in the development workflow. On top of that the construction is fairly verbose and debugging it is complicated as the noise specifications happen in various different places, instead of a central location.

Start by reading the QVMConnection API and the Device specification, focusing on how simple noise and full device definitions are currently supplied. Done means a coherent API supports advanced device-level noise, including non-native gates such as CNOT, without requiring major program rewrites.