Adding stochastic loss network element that randomly drops packets

[deeptir18]

-This element, similar to Delay, Link, and Receiver, has a:

• tick() function
• next_event_time() function
• accept() function

All it does is maintain a queue of packets to pass on to the next element, but uses a random seed to either drop or keep a packet before accepting the packet. I'm not 100 % sure that the next_event_time() function is completely accurate. The default loss_rate is set to 0, so all the regression tests should pass.

[keithw]

This looks really good! Couple of questions:

• Do you think you can break this up into some smaller commits? (Basically I'd love to see the smallest atomic commits you can do. E.g. one that adds the element, and then another commit that uses the element in the network, then another commit that adds it to the protobufs, then another commit that lets you set the arguments from the command line.)
• Talk me through the reasoning behind having the ConfigRange just have one stochastic loss rate. Is this a placeholder for later? (Which is fine.) Or do you want to have a range of loss rates that a RemyCC will be optimized over?

[deeptir18]

(1) Sure I can easily break it up into small commits and redo the request.

(2) I was thinking about how Remy is used - should we treat stochastic loss like everyone of the other configurable network variables, so we can theoretically train over a range over networks with different stochastic loss rates? I guess that does make sense for some cases - you'd want to have a Remy resilient to different amounts of loss. My decision was slightly arbitrary.

[keithw]

This is the most beautifully explicated sequence of commits this project has ever received. :-) Bravo, and thanks for sticking with this!