Fix for save_expval truncation

[gadial]

Summary

Fixes small problems preventing the usage of save_expval truncation.

Fixes #2249

Details and comments

PR #2216 added a mechanism for truncating save_expval, removing qubits for which the pauli measurement operator is I, since those qubits are not needed for the simulation of the measurement. This is crucial for running the simulator on small circuits compiled on large backends, since the save_expval is extended to all the qubits in the backend, meaning we might have a 127-qubit save_expval for a circuit with only 12 active qubits (e.g. in the added test case).

However, the truncation code was not activated in practice due to the !qubitset_.empty() check performed prior to the execution of the code which excludes I qubits. Since the save_expval command is the last one in the circuit when, e.g., running via an estimator (which adds the command to the circuit just before its run) and gates are traversed in reverse order at this stage, the save_expval command would always fail this test, defaulting to the qubitset_.insert(op.qubits.begin(), op.qubits.end()); line which adds all its qubits.

This PR removes the !qubitset_.empty() check and fixes some problems arising from this:

1. In edge cases, e.g. a circuit consisting only of a measurement of the pauli op I the truncation stage results in a circuit without qubits, and so it is skipped and no expval result it obtained. Hence, we always add the a qubit to the qubit set if it is empty, even if it corresponds to I. This does not enlarge the qubit set enough to pose a real problem, although smarter logics can be implemented.
2. A op.qubits.resize(qubitmap_.size()); command was missing in the remapping code (although present in the non-remapping truncation code in set_params); it was added. Without resizing, the save_expval command fails due to generation of invalid pauli mask data in the expval_pauli method.
3. When truncating the pauli string, the index of each pauli in the string is computed from the qubit index. However, in cases where the operator is only on a subset of the qubits, the qubit index might be too large. For example, if the pauli ZZ is used on qubits 1,2 and qubit 0 is truncated away, while creating the new pauli the code will attempt to access the pauli in index 2 (more precisely, in location pauli.size()-1-2 since paulis are accessed in reverse order). This is fixed by using the index of the qubits inside op.qubits instead of their absolute index.

[hhorii]

LGTM