Update to Cirq 0.13

[mpharrigan]

I'll track the effort needed to support Cirq 0.13 here.

• Lots of broken quantum chess tests: circuit_transformer.py:286 fails when it encounters 3-qubit gates. I think quantum_board.py:247 SycamoreDecomposer is not decomposing the tofflis @dstrain115
• Need openfermion that supports 0.13 Release version that supports Cirq 0.13 OpenFermion#751
• Verify latest pytket isn't over-pinned
• [pytket-cirq] Consider depending on specific subpackages CQCL/pytket-extensions#230 try to get pytket-cirq off of the cirq metapackage.
• Manually check notebooks

Other than quantum chess toffli check, all the tests pass.

[dstrain115]

Thanks for starting this effort. I will take a look at the quantum chess bug(s) early next week.

[mpharrigan]

likely quantumlib/Cirq#4167

[mpharrigan]

diff --git a/recirq/quantum_chess/circuit_transformer.py b/recirq/quantum_chess/circuit_transformer.py
index 4f3ff2b..6b999c4 100644
--- a/recirq/quantum_chess/circuit_transformer.py
+++ b/recirq/quantum_chess/circuit_transformer.py
@@ -415,7 +415,7 @@ class SycamoreDecomposer(cirq.PointOptimizer):
         new_ops = None
         if op.gate == cirq.SWAP or op.gate == cirq.CNOT:
             new_ops = cirq.google.optimized_for_sycamore(cirq.Circuit(op))
-        if isinstance(op, cirq.ControlledOperation):
+        elif isinstance(op, cirq.ControlledOperation):
             if not all(v == 1 for values in op.control_values for v in values):
                 raise DeviceMappingError(f"0-controlled ops not yet supported: {op}")
             qubits = op.sub_operation.qubits
@@ -440,6 +440,10 @@ class SycamoreDecomposer(cirq.PointOptimizer):
                     new_ops = cirq.google.optimized_for_sycamore(
                         cirq.Circuit(cirq.TOFFOLI(*op.controls, *qubits))
                     )
+        elif op.gate == cirq.TOFFOLI:
+            new_ops = cirq.google.optimized_for_sycamore(
+                cirq.Circuit(op)
+            )
         if new_ops:
             return cirq.PointOptimizationSummary(
                 clear_span=1, clear_qubits=op.qubits, new_operations=new_ops

fixes it in an expedient way