Support Numpy 2 via supporting Cirq 1.5+

[mpharrigan]

Changes for numpy

• data_types.py -- np.uint8 is stricter now
• and_bloq.by -- there were np.uint8's sneaking in
• There are new mypy complaints that I've generally ignored

Changes for Cirq

We require a more recent version of Cirq to support Numpy 2.

ControlledGate and _has_unitary_

I've monkeypatched cirq.ControlledGate. It was using cirq.has_unitary to determine whether something could be controlled. The rest of Cirq now treats "has unitary" as the ability to create a unitary matrix rather than saying whether the gate is reversible.

• Tracked in Reliance on cirq.ControlledGate is brittle #1764 .

Simulation

There is something fundamentally incompatible between Qualtran's cirq interop functionality and recent Cirq versions. I've taken the "nuclear option", and now when you request the cirq simulation of a BloqAsCirqGate it will be done entirely in Qualtran and export a large unitary matrix to the cirq simulator; which can cause performance regressions;

• see Cirq simulation regressions #1763.

In this PR, I've implemented the quick fixes to speed up Qualtran simulation of narrow, deep circuits from my comment on #1336

Some of the unit tests now cause SIGKILL on my machine. I don't know what's going on here. I've xfailed them.

• See Cirq simulation regressions #1763.

Related issues

• Implement catalytic controlled-Hadamard #237
• cirq.unitary fails for bloq with And followed by And^dagger #1488 (technically fixes, as written; but see performance regression issue)
• cirq.unitary does not work for controlled QubitizationWalkOperator operator #1495
• Tensor contraction 4-5x slower than cirq.unitary for hamiltonian simulation bloq #1336

[mpharrigan]

I've done local testing of the full unit test suite (not just the not slow tests) on mac and linux with Python 3.10 and newer, as well as with Cirq 1.6. The CI is still on Python 3.10 which caps us at Cirq 1.5. I'll leave bumping the CI Python version to a follow-up. So I think the nightly tests should pass; but we can't know for sure until this is merged. This is an area where the quick tests cannot catch all the errors that seem to happen only on the larger simulation tests.

[mpharrigan]

This broke the nightly tests, but there's no error message and I can't reproduce it