[FEA] Check that thread active in cuDF is holding onto the semaphore in the tests

[abellina]

With the work we are doing #7252 we need to make sure we are consistently grabbing the semaphore before calling into cuDF (and also when we are holding non-spillable memory).

We would like to add assertions that trigger in the tests where if a thread is known to be doing work on the GPU that it must be holding onto the semaphore, otherwise the concurrentGpuTasks config doesn't work in all cases.

This came about because of #7729, where the semaphore debug logic caught multiple threads doing exactly this.