`Y^-0.5` not being converted to `PhasedX(-0.5)^0.5` on Ion related tests in Windows.

As a reference: https://travis-ci.com/quantumlib/Cirq/jobs/201514754

For test_convert_to_ion_circuit

E       AssertionError: Circuit's text diagram differs from the desired diagram.
E
E       Diagram of actual circuit:
E       (0, 0): ───X───────MS(0.25π)───X^-0.5──────────MS(0.25π)───
E                          │                           │
E       (0, 1): ───Y^0.5───MS(0.25π)───Y^-0.5───S^-1───MS(0.25π)───
E
E       Desired text diagram:
E       (0, 0): ───X───────MS(0.25π)───X^-0.5─────────────────────MS(0.25π)───
E                          │                                      │
E       (0, 1): ───Y^0.5───MS(0.25π)───PhasedX(-0.5)^0.5───S^-1───MS(0.25π)───
E
E       Highlighted differences:
E       (0, 0): ───X───────MS(0.25π)───X^-0.5──────────█████████──████████████
E                          │                           ████████████
E       (0, 1): ───Y^0.5───MS(0.25π)───█████████████████████████──████████████

For test_decomposition

E       AssertionError: Circuit's text diagram differs from the desired diagram.
E
E       Diagram of actual circuit:
E       0: ───Y^0.5───Z───MS(0.25π)───Y^-0.5───S^-1───
E                         │
E       1: ───────────────MS(0.25π)───X^-0.5──────────
E
E       Desired text diagram:
E       0: ───Y^0.5───Z───MS(0.25π)───PhasedX(-0.5)^0.5───S^-1───
E                         │
E       1: ───────────────MS(0.25π)───X^-0.5─────────────────────
E
E       Highlighted differences:
E       0: ───Y^0.5───Z───MS(0.25π)───███████████████████████████
E                         │
E       1: ───────────────MS(0.25π)───X^-0.5──────────███████████

CC: @mlihc15 , @c-poole

[c-poole]

If you try to construct a PhasedX(-0.5)^0.5 cirq will recognize that as a Y^-0.5 and return that instead, but this relies on the -0.5 in the PhasedX passing a float equality comparison, so it isn’t too terribly surprising that floating point comparisons are causing system dependent differences. What seems to matter in this test is whether the circuits produced are equivalent and whether they pass the validate_circuit method of ion_device. I rewrote the tests with this in mind and they pass so I’ll put in a PR for it.

[vtomole]

@mlihc15 Yeah, this probably doesn’t have to do with the rendering because the only problem we seem to be having here is Y^-0.5 not being converted to PhasedX(-0.5)^0.5. This seems to be related to Windows somehow.