Mad

- Bug Fixes:
- Fix in the calculation of the throughput and reliability that leaded to negative values
- Fix the worker behaviour when request have timeout. They now check whether the task is still pending before to proceed
- Fix the invoke and query which now wait for acceptance by the server.
- Fix the behaviour of the invocation, which take 1 unit of time to send the requests
- Fix autoscalling issues that exceeds its limits by one.
- Fix R Script ylim which are now based the range for all related series

- MAD language:
- Support of an 'ignore-error' primitive
- Monitoring:
- overall response time
- per operation
- response time
- arrival rate
- reliability
- Bug Fixes:
- Fix R script to render 'missing values' (i.e., "NA") properly
- Fix multiple folders created when the simulation takes a long time
- Fix the validation so that it does not detect false 'never invoked operation' (on forward references)
- Fix R script to avoid plotting measures when all are NA (possible on edge cases)

- Support for retry on error, with back-off policies (constant or exponential)
- Support for transmission latency
- Support for service differentiation (unfair priority schemes)
- Support for timeouts
- Support for Probabilistic failure
- Reporting:
- reliability
- arrival rate
- throughput
- rejection rate, instead of rejection count
- Syntax:
- **IMPORTANT**: Using curly brackets to delimit blocks
- Bugs fixes
- Adding support for "missing values" in monitoring
- Refactoring
- Monitoring is now a separate object in the environment
- Throttling is now a decorator on TaskPool
- Logging is now isolated into wrappers

This version brings, among other things:
- One throttling strategy tail-drop
- Fix minor issues in the rule-based autoscaling
- Better error reporting and the static validation of the MAD models, covering:
- Duplicate identifier (for services and clients)
- Operation never invoked
- Operation invoked, but that are not defined
- Service that are not involved

This new version adds:
- Support for alternatives queue disciplines, LIFO and FIFO
- Supoort for rule-based autoscaling policies (acting on utilisation)
- Sketch of an emacs mode to edit MAD files

This new version includes a better scheduling engines that leverage closures in Python as well as a proper environment model, that ties together all the simulation entities.