MPS: Improve performance of Pauli expectation value

[merav-aharoni]

Summary

We improve the performance of computing the Pauli expectation value in MPS by inserting identity matrices on qubits that don't participate in the expectation value.

Details and comments

We implemented idea (2) suggested in issue #748.
When computing the expectation value on a subset of the qubits, in the previous implementation, we did the following:

1. Copied the MPS structure to a temporary structure
2. We centralized the relevant qubit using swaps, so that we could compute the exp. value on them only.
   In the new algorithm, we take the minimum and maximum qubits that participate in the exp. value, and compute the exp. value on all the qubits in this range, putting identity matrices on those qubits that do not need to participate.
   The new algorithm does not require a copy of the MPS and does not require swaps, and is therefore more efficient.

[chriseclectic]

What sort of performance improvement does this give?

[merav-aharoni]

[chriseclectic]

Looks like a nice performance improvement!