Better error message for bad loss function

[MilesCranmer]

When I try precomputing data rankings and then fitting a model using [incorrect API]

loss_function = """
begin
       using Statistics
       function loss_function(prediction, target)
           return 1 - cor(prediction, target)
       end
   end
"""

I run into the very long Julia error

Originally posted by @Waswas321123 in #981 (reply in thread)

Would be nice if the error message was better here. Rather than a MethodError we could
test the loss function earlier and validate it. Especially for Python users who might not understand it.