[Performance, Refactor] Operator multiplication without explicit identities

[jpmoutinho]

Refactors the operator_product function that does not require explicit identity padding. Also refactors the apply_density_mat into a new apply_operator_dm without explicit identities. The logic of both is similar, and the logic of apply_operator_dm is very similar to apply_operator_permute, but keeping everything separate for now.

The reason not to use operator_product inside apply_operator_dm is to avoid permuting the qubits twice.

The new functions are similar in performance for small number of qubits. For 5 to 9 qubits they become a few times faster. For 12 qubits the speedup is already ~10x.

[jpmoutinho]

Hey @EthanObadia finally got around to writing the apply_operator equivalent for density matrices without using explicit identity padding. If you have some time to check it out it would be great :)