Allow protocols to work on raw arrays

[daxfohl]

Is your feature request related to a use case or problem? Please describe.

When working on #7048, was surprised that unitary(val) and apply_unitary(val, args) didn't accept raw numpy unitaries for val. This made it necessary to special case a couple isinstance(np.ndarray) checks in the consumers of those functions. There is also an already-existing redundant function in apply_mixture that can be removed if apply_unitary can accept raw numpy unitaries. If accepted, this could be a good first issue for someone.

Describe the solution you'd like

I started this for apply_unitary in #7039, but closed the PR because a) I think it's a maintainer decision as to whether the feature is desired, how large the scope should be, and whether it can be delivered incrementally or should go all-at-once, and b) I think it would be a good first issue for someone to learn about quantum operators in general, and cirq's implementation of protocols in particular.

As mentioned in the closing comment #7039 (comment), the scope is open-ended.

Closing for now. Seems like if we do this, then we should also do unitary and has_unitary. Which, then may as well do mixture and kraus series, maybe act_on, and whatever else comes up too. Support raw Python arrays? Sympy?....

[optional] Describe alternatives/workarounds you've considered

Duplicate code, instanceof checks.

[optional] Additional context (e.g. screenshots)

As part of this, I noticed the protocols have different techniques. Some loop through an array of strategies, others are more imperative. Some use dynamic invocations of the fallback strategies, whereas others call the fallback protocols directly. It'd be good to improve consistency in these when working through them.

What is the urgency from your perspective for this issue? Is it blocking important work?

P3 - I'm not really blocked by it, it is a suggestion based on principle

[NikolaiLong]

Hello @daxfohl, I'm new to this project and have just set up my environment locally with all tests passing. I have many years of experience with python and I would like to work on this issue. Could you assign it to me please? Thank you!

[daxfohl]

I'm not a maintainer, and the issue is still in triage state. Feel free to do it, but there is no guarantee that the issue will be accepted. Maybe attend the Wednesday sync if you'd like to petition for it.

[NikolaiLong]

Thank you for the clarification. I'll attend Wednesday's sync.

[mhucka]

Discussed during Cirq Cynq 2025-03-05: accept as a worthwhile change. Limit the data types to things that can be turned into NumPy arrays. (Users will have to convert other things like SymPy objects to NumPy arrays if they need to do that.)