Provide a way to avoid copying state vector from Cirq

[jkalsi1]

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

See: quantumlib/qsim#893

Describe the solution you would prefer

Implement a flag to allow/disallow state vector copy

How urgent is this for you? Is it blocking important work?

P3 – I'm not really blocked by it; it's an idea I'm proposing based on principle

Describe alternatives/workarounds you've considered

Provide additional information or context (e.g., screenshots)

[jkalsi1]

@mhucka after your review, may I be assigned to this issue please?

[dstrain115]

Temporarily assigning to @mhucka and @sergeisakov to review the proposed flag solution (see attached qsim issue: quantumlib/qsim#893)

[sergeisakov]

I think that the proposed flag solution to toggle the initial state copy and buffer creation is a good temporary workaround. It provides immediate relief for out of memory issues when running large circuits in qsimcirq without requiring a massive architectural change right away.

However, I think that it should not be a permanent fix. The root cause is a design mismatch: a state designed for internal simulation buffering is being repurposed as the final simulation result.

Moving forward, I would suggest decoupling the Cirq's internal buffered state from the final StateVectorTrialResult and ensuring that qsimcirq can return the state vector directly from the C++ backend to the user without hitting Cirq's intermediate Python-side buffers.

Let us proceed with the flag solution for now, but I would suggest opening a separate issue to track the decoupling of the result state from the simulation buffer.

[jkalsi1]

Thank you for the proposed solution. In addition to copying the initial state, _BufferedStateVector creates a buffer in init. The buffer creation should also be toggled by this flag.

Copying from @sergeisakov in quantumlib/qsim#893 to capture in task scope.