Add isolated qubit functionality for gridmetadata.

[MichaelBroughton]

Adds isolated qubit property to GridDeviceMetadata. Also adds error checking for self loops. cc: @verult .

[mpharrigan]

did you consider having the arguments be qubit_paris, all_qubits and have union semantics rather than the current intersection semantics? Networkx has more of the former when you can

g = nx.Graph()
g.add_edges_from(edgelist)
g.add_nodes_from(nodelist that may or may not include nodes in edges)

[verult]

What if we default isolated_qubits to an empty collection instead of None?

[MichaelBroughton]

Using empty collections is a dangerous default.

Here's a simplified example:

def f(value, key, store={}):
    store[value] = key
    return store

print(f('a', 1))
print(f('b', 2))

prints:

{'a': 1}
{'a': 1, 'b': 2}

Using it as None and then promoting it is a safer route that's recommended in the style guide.

[verult]

WDYT about using all_qubits instead?

Pros

• DeviceSpecification keeps a full list of qubits, so the translation is easier. Current design plans to keep this field. This will minimize changes on Pyle side.
• It will simplify the logic of computing the node set a bit, and also computing all_qubits for the super class constructor.

Cons

• If we want to expose an isolated_qubits property, still need to calculate that separately. This somewhat voids the benefit of reducing code for computing all_qubits.

[verult]

Just saw @mpharrigan 's comment :P

[MichaelBroughton]

Changed over to all_qubits approach. PTAL.

[verult]

Thanks for explaining the None default! LGTM

[verult]

nit: can be in the else block of if all_qubits is None.

[MichaelBroughton]

Oops looks like @verult isn't a code owner for cirq-core. We should probably fix that. In the meantime, would @95-martin-orion you mind approving as well ?