Using numpy API from rust changes how threading works

[clbarnes]

Previously, users had fine-grained control over the number of threads to use on every query.

In the process of converting ncollpyde to be able to read arguments directly from numpy's buffers #1 (which significantly improves speed and memory profile), this fine-grained control was lost: queries either happen in serial or using the global rayon thread pool. This additionally saves on the overhead of building a thread pool was built for every call to a Volume method.

You can set the number of threads in the rayon thread pool by setting the environment variable RAYON_NUM_THREADS before the global pool is built (probably on the first query, but setting it before ncollpyde is imported may be safest). By default, the number of logical CPUs is used. More information is available here. You can find out how many threads are being used with the ncollpyde.n_threads() function.

ncollpyde's API will remain unchanged for the moment, but deprecation warnings will be thrown when non-boolean values are given.