What should we add?

Currently the hashing of a Pauli operator is computed as hash(self.to_label()). It is basically asking for the label of the Pauli operator which is an expensive computation to run. Instead building it directly from self.x, self.z,self.phase seems more reasonable.
I did a quick test where I do the hashing with int.from_bytes(np.packbits([self.x, self.z]+[self.phase])) and it was about 50 times faster.