Inconsistent behavior with mid-circuit measurements while sampling if NO reset is performed after the measurement

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

Currently, the presence of mid-circuit measurements has no "collapsing" effect on most of our simulators. This leads to an inconsistency in results when we do shot-by-shot versions of sample (due to the presence of if checks in the code) vs when we do a single pass and sample once at the end.

For simulators, mid-circuit measurements should collapse the state (statevector or density matrix, etc.), but they do not currently do that.

Steps to reproduce the bug

Run the following test kernels, which are semantically identical to each other:

import cudaq

@cudaq.kernel
def midCircMeasTest1():
    q = cudaq.qvector(1)
    h(q)
    midCircuit = mz(q)
    h(q)

@cudaq.kernel
def midCircMeasTest2():
    q = cudaq.qvector(1)
    h(q)
    midCircuit = mz(q)
    # Put a dummy if check here to force shot-by-shot sampling
    if midCircuit:
        pass
    h(q)


print('midCircMeasTest1 results')
print(cudaq.sample(midCircMeasTest1))

print('midCircMeasTest2 results')
print(cudaq.sample(midCircMeasTest2))

$ python3 test.py
midCircMeasTest1 results
{
  __global__ : { 0:1000 }
   midCircuit : { 0:513 1:487 }
}

midCircMeasTest2 results
{
  __global__ : { 0:482 1:518 }
   midCircuit : { 1:497 0:503 }
}

The problem exists for all of our simulators except for stim.

Expected behavior

At a minimum, the two example kernels should produce similar results in the __global__ register. And it seems like the shot-by-shot implementation is more correct right now (showing a 50/50 split on the __global__ register), which is what would happen if you ran this on real hardware.

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

Not a regression

Environment

• CUDA Quantum version: latest (4e069429bbc7c8715c053ee32b8f7c29ce10276b)
• Python version: 3.10
• C++ compiler: N/A
• Operating system: Ubuntu: 22.04

Suggestions

We likely need to make sure that our simulators collapse the state after sampling. For optimality, we probably only want to do this on mid-circuit measurements (right before executing the next gate), not final measurements (where there is no next gate to worry about).