Ssbench

* Fixed 23: Allow per-size CRUD profile in scenarios.

* Fixed object uploading bug when size > several chunk sizes with retries

* Fixed 27: Replaced eventlet with gevent to improve HTTPS support

* Fixed bug where "." was printed during a run for results with no
first_byte_latency (i.e. CREATE or UPDATE) instead of "_".

* Fixed 15: added CHANGELOG and AUTHORS; added "Contributing to ssbench"
section to README.

* Cleaned up PEP8 compliance and added .pep8 to ignore E501: line too long

* Added ssbench-master scenario op count override.

* Fixed up docs (README.rst).

* Fixed 13: can set ssbench-worker concurrency (again)

* Removed stale dependency on PyYAML.

* Fixed 11: added --noop mode.

* Fixed 5: replaced YAML with msgpack.

* Fixed 12: addded ability to profile master and worker.

* Got rid of some result-sinking code which was slowing the master down
quite a bit.

* Fixed 4: replace median with configurable percentile.

* Fixed 1: handle zero responses per CRUD type.

* Fixed 2: add start/stop times & duration to report.

* Fixed 3: add timestamp to default output stats filename.

* Fixed bug in reporting where per-size stats per operation type were using
the "all sizes" data.

* Fixed bug in GET where last_byte_latency was being calculated before
actually reading the data out of the socket.

* Some operations cannot get a meaningful first-byte-latency, so they
don't; this is now correctly reported as "N/A".

* Each result's Swift X-Trans-Id header is now included in the result.

* Added "Swift TX ID for worst latench" column to report.

* Increased precision in report data (now visible down to milliseconds).