Regression for cirq.GridQubit with float indices

[dstrain115]

Description of the issue

#6987 coerces GridQubit indices to be ints.
If a float is specified, then the behavior changes from the most recent version.

How to reproduce the issue

cirq.GridQubit(4.5, 2.5)

Details

Before, this would return cirq.GridQubit(4.5, 2.5) and now returns cirq.GridQubit(4, 2)

Cirq version
dev1.5.0

[dstrain115]

@daxfohl Do you remember why you needed to convert to int in #6987?

Is it possible to do something less extreme to preserve behavior (at least for now)? For instance, can we type cast to int instead or coerce to int only if it is a numpy type?

I guess the QEC folks like to use half-integers for QEC circuits since it is convenient when using stim (ie. data/measure qubits etc)

[daxfohl]

Will look into it. I think it was because hash needs to return a python int, and returning a numpy int failed. If it needs to support non integral types, then that is moot. We can call hash(row) and hash(col), which will give us back the same int for ints, and a valid hash for non ints. We'll see.

[pavoljuhas]

We can call hash(row) and hash(col) ...

Or consider just wrapping the current expression in hash(...).
On a quick check hash(3) == hash(np.int32(3)) == hash(np.int64(3)).

Also, we should address this for the LineQubit too.