Marginalize probabilities prior to confusion matrix application

[alecandido]

Since the readout confusion matrices applied are only defined per-qubit, they do not mix the probabilities of the states of the other qubits.
Moreover, since we do only support measurements for individual qubits, we can then safely marginalize on all the other qubits spaces prior to the application of the confusion matrix for the single qubit involved in the measurement.

For an individual array of probabilities, the code would look like the following

dims = (3, 3, 3, 3)
p = np.random.rand(np.prod(dims))
probs = p / p.sum()

def marginalize(p: np.ndarray, idx: int, dims: tuple[int, ...]) -> np.ndarray:
    return p.reshape(dims).sum(axis=tuple(i for i in range(len(dims)) if i != idx))

x = marginalize(probs, 2, dims)

Pay attention that this needs to be generalized when applied simultaneously for multiple distributions (as in the results of swept experiments.

Note

We only started applying confusion matrices in #1401