radosbench: support total_threads parameter as alternative to concurrent_ops

[tdeneau]

This pull request adds a total_threads parameter to the radosbench benchmark. which is an alternative to concurrent_ops (concurrent_ops behavior is not changed). If total_threads is used, the framework will calculate the concurrent_ops needed (if any) for each of the concurrent_procs so that the total will match total_threads.

Also we add concurrent_procs to the archive directory structure so it can be treated like other list values in the yaml file if desired.