Job pool hangs on pool.join with maxtasksperchild set

[MatthewThe]

Setting maxtasksperchild results in the pool hanging on pool.join.
The problem seems to be that the function called asynchronously uses some matrix manipulation functions that numpy parallelizes with OMP.

By prepending the command with OMP_NUM_THREADS=1, the issue is resolved.

Using threadpoolctl we could achieve this programmatically. This either has to happen in job-pool or in the program calling job-pool.

[MatthewThe]

Setting the number of OMP threads with threadpoolctl seems to help, but the pool.join still hangs every now and then.