Fix local_pauli twirling with Clifford and non-Clifford fractional entanglers in the same box

[Joshua Skanes-Norman (joshuasn)]

Summary

When using local_pauli on a box with a Clifford and non-Clifford angle, e.g.,

with c.box([Twirl(group="local_pauli")]):
    c.rzz(np.pi / 2, 0, 1)
    c.rzz(np.pi / 3, 2, 3)

with c.box([Twirl(group="local_pauli")]):
    c.rzz(np.pi / 2, 0, 1)
    c.rzz(Parameter("a"), 2, 3)

the emits sample from the Pauli group when the angle is Clifford but the corresponding propagate ends up using PropagateLocalPauli, which guards against Paulis not in its commutant. This caused the SamplexRuntimeError.

This PR updates the logic to dispatch to the correct propagator.

[Ian Hincks (ihincks)]

This change does look to fix the error at hand.

There is a bigger question about whether local_pauli should twirl with the full pauli group at the special rzz angles 0 and pi/2, but that's out of scope for now: we need to fix the broken path.