Negmas

---------------

* Tournament combination in Cartesian tournaments

* Cartesian tournaments were structured to allow for saving and loading
* Added load/save/combine to SimpleTournamentResults
* Added combine_tournament for combining on disk tournament results with control on copying and scenario renaming

* Adding max-tasks-per-child to all tournaments
* Supporting world-timeout in tournaments
* Do not end tournaments if a single future timesout

---------------

* Saving repeated runs in cartesian tournaments.
* Adding sort_runs to cartesian tournaments. Passing sort_runs overrides randomize_runs and causes smaller scenarios to be negotiated first. Scenario's size is the product of the outcome-space cardinality, number of steps, and time-limits.
* Improving parallelism in cartesian tournaments:

* Adding max-tasks-per-process
* Adding external_timeout to cartesian tournaments. This parameter is only effective for parallel runs and enforces and external timeout on all tasks. This will avoid waiting for slow negotiators but it has the disadvantage of causing the main thread to never end. Tournaments may now need C-c to end.
* Adding _InfiniteWaiter negotiator for testing
* Accepting None for hidden_time_limit in Mechanism.

---------------

* More exception resistance in cartesian_tournament
* Base Mechanism class now can ignore exceptions in initialization of agents
* More consistent naming of methods involving negotiators

---------------

* Matching the behavior of Genius in relative_time.
Originally, NegMAS used the following rule for relative_time R:

i = 0 => R = 0

i > 0 => R = min(1, max(t/T, (i+1)/(N+1)))

where i is the current step, t is current time, T is time_limit, N is
n_steps.

After this change the special case at i=0 is removed:

R = min(1, max(t/T, (i+1)/(N+1)))

This matches Genius behavior exactly if we have either T=\infty or
N=\infty.
* More info in printing in cartesian verbose
* Rounding output in cartesian verbose mode
* Printing execution time in cartesian verbose

---------------

* Saving erred-negotiator and erred-agent in MechanismState in case of errors in SAO and related mechanisms
* Saving details about errors (including traceback) in cartesian products (in all_scores.csv).

---------------

* Improving the way some stats are saved in situated
* CLI improvements

Now the "negotiate" CLI has a much better help. Try:

>> negotiate --help

Moreover, we added the ability to generate scenarios and save them as
well as more control over statistics