There has been no activity on issue https://github.com/numba/numba/issues/9103.
This means that all experiments using numpy's random.Generator as parameter of a numba-annotated function will crash.
They will crash due to a memory leak in numba and they will crash without notice or output, often taking the bash shell down with them.
Therefore, for now, all the annotations of the kind
numba.njit(cache=True, inline="always", fastmath=True, boundscheck=False)
before the search operators have been removed.
We will take the performance hit here and wait until the issue is resolved.