Loss curves and units

[luke-kiernan]

Currently, fields that are loss curves are declared as ValueCurve types. Unlike CostCurve types, those objects don't have units data attached to them. Hence the units are currently ambiguous.

This is causing actual issues. psi_library.jl:1829-1957 in PSB builds a loss curve on an InterconnectingConverter with the y axis in system base. Meanwhile, code in an open POM PR assumes that same curve has its y axis in device base...

Potential solutions:

1. Eliminate the ambiguity by adding units somewhere.
   a. Add units to ValueCurve.
   b. Introduce some other curve type that carries units, but isn't a CostCurve.
   c. Just make them CostCurves, even though losses aren't costs.
2. Decide on some fixed units and document that choice prominently.

edit: here I'm talking about the IS4/psy6 branches.

[luke-kiernan]

Of all the options above, (1b) seems the least bad. @jd-lara unless you have strong opinions otherwise, I'll go implement that.

[jd-lara]

Let me think about what the best option is here.

[luke-kiernan]

Here's all the loss curves:

• TwoTerminal{GenericHVDC, LCC}Line: loss::Union{LinearCurve, PiecewiseIncrementalCurve}
• InterconnectingConverter: loss_function::Union{LinearCurve, QuadraticCurve}
• TwoTerminalVSCLine: converter_loss_{from, to}::Union{LinearCurve, QuadraticCurve}

Here's my proposal: rename ProductionVariableCostCurve to ProductionVariableCurve (or similar), then introduce a new type LossCurve{T, U} <: ProductionVariableCurve{T, U}. Here T is curve type (quadratic, linear, or piecewise linear) and U is units. If it's important that e.g. LCC's can't have quadratic loss curves, then we add a validation step in set_loss!, at the cost of moving those functions from generated to hand-written.