Merge branch 'segmented-threads'

Pull in the segmented segments code. This avoids having to setup a
thread area segment for the max (4096) number of jobs. Instead it's
done in chunks of 8. This results in a net reduction of more than 1G
of memory used for the normal use case.

* segmented-threads:
  Kill off 'max_jobs'
  Add thread_segments as needed
  Wrap thread_data in thread_segment