Custom loss functions - v0.5.0

[MilesCranmer]

This will move PySR to v0.5.0 of SymbolicRegression.jl (once that version is added to the registry).

This also enables custom loss functions via the loss kwarg. One can define a loss using the functions available (e.g., L2DistLoss() or L1DistLoss()), or pass a function definition like this:

pysr(..., loss="myloss(x, y) = abs(x - y)")

or, for weighted, e.g.,

pysr(..., loss="myloss(x, y, w) = abs(x - y) * w")

This is slower than using the built-in losses but more convenient for defining new ones.