Carpet: Add timing::total_threads variable

[rhaas80]

@ianhinder's pull request 20 from https://bitbucket.org/eschnett/carpet/pull-requests/20

I would like to be able to analyse output data and determine things like computational cost, and number of cores used as a function of time. This branch has a prototype which satisfies my immediate needs, but might not be the best approach.

• It would also be nice to have directly the computational cost, which would probably be time_total * total_threads. This would be the cost for the current Cactus segment.
• Maybe we want cores rather than threads
• Maybe we want total cores available on the node rather than threads, since this is what is “spent”, even if it is not being used.

Thoughts?

[rhaas80]

Ian, Erik, do you want to champion this? My personal opinions is that I would rather not output this static information at each timestep along with the other timing information though realistically, putting it eg in an extra Cactus group to request output for it less frequently is not going to make much of a difference (plus the number can change in between restarts after all).

[rhaas80]

Maybe also add the number of MPI ranks (for the same reason)? Though I would normally expect such static information to be written out only once and not each timestep.