Noisy simulation on tensornet-mps backend is not deterministic when manually setting random seed

Required prerequisites

• Consult the security policy. If reporting a security vulnerability, do not report the bug using this form. Use the process described in the policy to report the issue.
• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker to verify that this hasn't already been reported. +1 or comment there if it has.
• If possible, make a PR with a failing test to give us a starting point to work on!

Describe the bug

When repeatedly doing a noisy simulation using the tensornet-mps backend on the same quantum circuit C with the same random seed r, the sampling results are different. This shows that setting a random seed does not guarantee that the simulation is deterministic and reproducable.

Steps to reproduce the bug

In order to reproduce the bug you just need to create a simple kernel that applies x gate on a qubit. Add a DepolarizationChannel to the x gate on said qubit, set a random seed and sample the kernel. You will observe that consecutive runs of this program will give different results/histograms. I have attached a sample program to reproduce this issue. I used a for loop in my example, but same issue occurs when you just sample once and rerun the python file.

import cudaq

cudaq.set_target("tensornet-mps")


@cudaq.kernel
def my_kernel():
    qubits = cudaq.qvector(1)
    x(qubits[0])
    mz(qubits)


noise_model = cudaq.NoiseModel()
noise_model.add_channel("x", [0], cudaq.DepolarizationChannel(0.1))

for i in range(5):
    cudaq.set_random_seed(42)
    result = cudaq.sample(my_kernel, shots_count=1000, noise_model=noise_model)
    print(result)

Expected behavior

Expected behaviour would be that we get the same sampling result 5 times. Instead, we get this:

{ 0:56 1:944 }
{ 0:72 1:928 }
{ 0:57 1:943 }
{ 0:67 1:933 }
{ 0:59 1:941 }

Is this a regression? If it is, put the last known working version (or commit) here.

Not a regression

Environment

• CUDA-Q version: 0.11.0
• Python version: 3.12.9
• C++ compiler: -
• Operating system: Ubuntu 22.04.5 LTS on WSL

Suggestions

No response

Start with the reproducible Python entry points: cudaq.set_target("tensornet-mps"), cudaq.set_random_seed, and cudaq.sample using the provided noisy kernel. Run the five-iteration example and compare the histograms across repeated executions. Done means repeated sampling with the same seed produces the same results on the tensornet-mps backend, with coverage for the reproduced case.