Fix sampling with mixed-basis measurements (mx/my)

[anpaz]

Fixes #4333

Root Cause

In sampling mode, mx(q) is implemented as h(q); mz(q) — a basis-change gate followed by a Z-basis measurement. The problem is that every gate application calls flushAnySamplingTasks(), which samples all previously recorded qubits and clears the measurement buffer (sampleQubits). So the h gate inside mx prematurely flushes measurements recorded before it, and only the qubits measured after the flush end up in the final result.

For example, in a kernel that does mz(q[0]); mx(q[1]):

1. mz(q[0]) records qubit 0 in sampleQubits
2. mx(q[1]) calls h(q[1]), which triggers flushAnySamplingTasks() — this samples qubit 0 and clears sampleQubits
3. mz(q[1]) (inside mx) records qubit 1 in sampleQubits
4. At finalization, only qubit 1 is in sampleQubits, producing 1-bit results instead of 2-bit

Fix

Skip the non-forced flush in flushAnySamplingTasks() when in sampling mode without conditional measurements. Per the spec, in non-explicit sampling mode, qubit values correspond to the state at the end of kernel execution — the final forced flush in finalizeExecutionContext handles the actual sampling. Mid-circuit flushes are only needed for conditional measurements (if mz(q)), which is already tracked by hasConditionalsOnMeasureResults.

Testing

Added regression tests (C++ and Python) that verify mixed-basis measurement kernels return correctly-sized bitstrings with proper qubit ordering.

[mitchdz]

Is this file being generated from somewhere new? If it's just local for you, add it to your .git/info/exclude

[anpaz]

Not sure. This folder appeared with test results. I ran the tests from the cli using ctest and pytest. I also triggered the tests from VSCode, so I'm not sure what exactly is creating the folder.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.