Add functionality to generate confusion matrix on backend

[natestemen]

In REM, we have a handy function to generate a confusion matrix when the probabilities of bitflips are known: https://github.com/unitaryfund/mitiq/blob/7984127cb57042bdcd02462d18aee32aa6c84444/mitiq/rem/inverse_confusion_matrix.py#L74-L78

Adding a function that generates the inversion confusion matrix when the parameters $p_0$ and $p_1$ are not known would aid users in using REM when using real backends with limited knowledge.

Requirements

1. Decide best way to "learn" $p_0$ and $p_1$ parameters, especially when they may differ across qubits.
2. Add a function to mitiq/rem/inverse_confusion_matrix.py that allows a user to generate an inverse confusion matrix on a specified backend (using an executor). This function should optionally return $p_0$ and $p_1$.

---

This idea came out of discussion with @ecarlander and @ruheenirodi.