More interesting OSX issues

I have a CI setup for tests on OSX for the ducc package, which contains an almost verbatim copy of pypocketfft's threading functionality. One thing I learned was that in certain circumstances the C++ compiler defines some functionality macros, despite the functionality not being available (see https://travis-ci.com/github/litebird/ducc/jobs/380712386).

Also it appears possible to create deadlocks with the new threading implementation on OSX (see the other recent failed builds at https://travis-ci.com/github/litebird/ducc/builds). It's quite possible that I broke something when I adjusted the threading code for ducc, but on the other hand the changes were quite minimal. @g-peterbell, if you have an idea what could be going wrong and/or think that the pypocketfft might also be vulnerable to this, please chime in!