Compile-time issues with update funcs

[gaurav-arya]

There are some surprising compile times when repeatedly creating an operator:

using SciMLOperators

@time ScalarOperator(0.0; update_func=(u, p, t)->3) # 0.02 secs
@time ScalarOperator(0.0; update_func=(u, p, t)->3) # still 0.02 secs
@time ScalarOperator(0.0; update_func=(u, p, t)->3) # still 0.02 secs

If this an issue, perhaps it's due to the FilterKwargs design -- we may want to consider tweaking the internal design to reduce compile time, perhaps using function wrappers?

[ChrisRackauckas]

Every anonymous function is a unique type, so that's going to recompile the things hitting the update_func type.