Substates are separated after measurements, when measurements are ignored

[daxfohl]

Description of the issue

Density matrix separates qubit states after measurement when split_untanged_states=True. However when ignore_measurement_results=True, this should not happen, as the this changes the measurement into a dephase and does not make the state separable.

For instance, when measuring a Bell state, the resulting DM should be 0.5 |00> + 0.5 |11>. However, separating those states (partial tracing each qubit) and re-kronning them gives 0.25 of each; i.e. it causes each qubit to be 0.5 |0> and 0.5 |1> independently. Therefore we need to avoid separating states after measurements if ignore_measurement_results=True.

How to reproduce the issue

def test_ignore_measurements_remains_entangled():
    q0, q1 = cirq.LineQubit.range(2)
    simulator1 = cirq.DensityMatrixSimulator(
        ignore_measurement_results=True, split_untangled_states=False
    )
    simulator2 = cirq.DensityMatrixSimulator(
        ignore_measurement_results=True, split_untangled_states=True
    )
    circuit = cirq.Circuit(
        cirq.H(q0),
        cirq.CX(q0, q1),
        cirq.measure(q0),
    )
    result1 = simulator1.simulate(circuit)
    result2 = simulator2.simulate(circuit)
    np.testing.assert_almost_equal(result2.final_density_matrix, result1.final_density_matrix)