Affinity support in Thor

From various pieces of testing, processor affinity can have a big difference on the speed of multi threaded systems.

We should consider having an option that would bind a slave process to a particular socket if there were more slaves than sockets enabled on the system. (Maybe another separate option to restrict to cores within the socket if there are > 1 processes per socket.) If we could also use the correctly associated memory that would provide more advantages.

The disadvantage is that sort (and other cpu intensive operations) wouldn't be able to use all the cpus. In practice that is likely to lead to overcommited cpus anyway.

The most efficient set up may possibly be 2 (number of socket) slave processes per node, and multiple slaves per slave process.