run_opencl_fft: NVIDIA CUDA queue check is inverted (raises on same queue)

[xywei]

Summary

In sumpy.tools.run_opencl_fft, the NVIDIA CUDA queue validation condition appears inverted.

Current code on main:

for evt in wait_for:
    if not evt.command_queue != queue:
        raise RuntimeError("Different queues not supported with NVIDIA CUDA")

not (evt.command_queue != queue) is equivalent to evt.command_queue == queue, so this raises for the same queue (the safe/expected case), and does not raise for different queues (the unsafe case).

Expected behavior

Raise only when an event in wait_for belongs to a different queue:

if evt.command_queue != queue:
    raise RuntimeError("Different queues not supported with NVIDIA CUDA")

Why this matters

On NVIDIA CUDA OpenCL, valid FFT calls with same-queue wait_for events fail with:

RuntimeError: Different queues not supported with NVIDIA CUDA

This also breaks downstream CUDA tests (example observed in volumential):

• test/test_volume_fmm.py::test_volume_fmm_laplace fails before the fix
• passes after changing the condition to evt.command_queue != queue

Additional context

This looks like a logic typo introduced in the NVIDIA marker handling path (Fix enqueue_marker for NVIDIA CUDA, commit 9d1d611), likely unintended.