Follow-up fixes for OpenBLAS-pthreads controlling thread feature

Document behavior when one-off API is used together with context managers (ownership is released)
Document NumPy and SciPy can have their own different OpenBLAS, but there's only one controlling thread.
Figure out if we ever leave a thread erroneously in control, which is more significant problem than losing ownership. (Tested joblib, doesn't leave anything in control. Thought about it and read code, doesn't seem likely.)

I will take this on.

Navigation Menu