ROSS/CODES sampling is a mess

[nmcglo]

I sort of mentioned this in #164 but feel like it needs its own issue as the other one's original text had been addressed.

The statistics that are put into the sampling structures is kind of a mess and isn't well documented. For example we have code like this in dragonfly-dally.C's packet_send() function:

        if(s->last_buf_full > 0.0)
        {
            bf->c6 = 1;
            msg->saved_total_time = s->busy_time;
            msg->saved_busy_time = s->last_buf_full;
            msg->saved_sample_time = s->busy_time_sample;

            s->busy_time += (tw_now(lp) - s->last_buf_full);
            s->busy_time_sample += (tw_now(lp) - s->last_buf_full);
            s->ross_sample.busy_time_sample += (tw_now(lp) - s->last_buf_full);
            msg->saved_busy_time_ross = s->busy_time_ross_sample;
            s->busy_time_ross_sample += (tw_now(lp) - s->last_buf_full);
            s->last_buf_full = 0.0;
        }

poor re-use of unrelated RC msg variable names aside, there's obviously a lot of repeated behavior encoded in this snippet. Why is there a s->ross_sample.busy_time_sample value AND a s->busy_time_ross_sample value? What is the difference?

[nmcglo]

@caitlinross, I'd love some time to chat if you had some insight about these features, I've assigned you on this issue just to consult (unless you wanted to tackle it all yourself!)

[caitlinross]

Whoops I'm not on github much (we use gitlab at Kitware) so I completely missed this. Probably got an email but I've gotten so many emails from the repo lately that I probably missed it. :)

Anyway, at some point after SC or maybe Thanksgiving I can probably take some time to help out.