move getModeledWork out of TemeperedWMin

[cz4rs]

Having getModeledWork baked into load balancer makes it impossible to compute post-LB statistics. This could possibly be moved into a new load model, but we have to find a way to pass parameters into it (alpha, beta, gamma).
A more general solutions for load models might be required.

---

Should this be implemented as a new load model instead of integrated into the load balancer? Load models can be composed, so you'd still be able to plug in whatever model you wanted for the loads.

Having the formula integrated into the LB makes it impossible for vt to compute post-LB statistics following the formula. What we dump to the LB statistics json file post-LB would be incorrect but I'm not sure if it would have any other immediate consequences. I think it might break the (not yet implemented) capability to run multiple load balancers and use the most effective result.

What do you think, @PhilMiller and @ppebay?

Originally posted by @nlslatt in #1695 (comment)

[cz4rs]

As discussed on LB meeting:

• implement WeightedCommunicationVolume load model that takes alpha, beta, gamma as ctor arguments

Possibly in another issue: move into #1884

• add labels to load models (including suitable mechanism for composed models!)
• include the labels in post-LB JSON output