For first order we can optimize the evaulation of the finite diff operator (which is something the DI operator in #88 ) currently doesnt do.
Additionally we should have a common interface for reuse of the primal evaluation to support things like #156
For first order we can optimize the evaulation of the finite diff operator (which is something the DI operator in #88 ) currently doesnt do.
Additionally we should have a common interface for reuse of the primal evaluation to support things like #156