`cnmfsns factorize` currently uses all available CPUs per worker instead of the CPU affinity or a --cpus parameter.

[verheytb]

The bottleneck to this is that paralellization is hard-coded in the cnmf module. To implement it, we will need to copy and modify that code.

[verheytb]

Now that the external cNMF dependency is resolved, we can implement this easily with threadpoolctl package.