Refactor iterative mle state tomography to use pyquil operator estimation

[mpharrigan]

iterative_mle_state_estimate's function signature has been updated to take a list of pyquil's ExperimentResults in place of TomographyData in #4. However, under the hood it just shims it into a TomographyData and does normal processing.

The processing needs to be modified to support ExperimentResults directly. This probably involves re-working _R to operate on expectation values rather than backing out counts and projectors. [Note that we pull this trick for linear inversion]. In general, this function should be ported to match the style of linear inversion:

• operate on ExperimentResults (in any order)
• return data should just be "new" stuff, like estimate of rho (+ log likelihood?)
• Use scipy's pinv
• Remove usage of construct_projection_operators_on_n_qubits / n_qubit_pauli_basis in favor of pyquil.unitary_tools.lifted_pauli and PauliTerms from ExperimentResults.

[joshcombes]

On the last point.

I think we should be careful about relying on the pyQuil unitary tools etc. For many reasons (e.g. the pyQuil re-org) and they don't seem to have informative names unlike construct_projection_operators_on_n_qubits. But I agree with the substantive part, i.e what Matt suggests is a good idea and we should certainly move in that direction.