Add extract timing hook

[rcthomas]

• Insert an optional timing=None argument. If this argument is a
  dict then it will be updated with timing information while the
  function is running.
• Mark points in time for various phases including read_input,
  preparation, total_extraction_time, total_output_time,
  and merge and compute elapsed time from differences.
• Marks and some timing computations are done whether or not any
  timing argument is passed, but it is only set if a dict is
  passed. Any other type of argument, the timing data is lost.

In the long run a more uniform strategy for collecting timing data
should be considered and these changes supplanted by that, but this
is good for present purposes of understanding where the code is
spending its time at various concurrencies.

[sbailey]

Looks good; thanks.