Adds a specialization of MOI.delete that deletes a batch of constraints.

[henriquebecker91]

I have only did it for MOI.ScalarAffineFunction constraints. This is merely a performance improvement. JuMP and MathOptInterface already have batch constraint deletion fallbacks and they will correctly call this method when it is the case. I did not add new tests nor benchmarks, but did call JuMP.delete in my code (that had performance problems), checked if the method was really called (with a println already removed), and the performance difference was from 23.2 seconds to 0.531 seconds (~29k constraints deleted from ~36k constraints, the speedup was probably higher than the ~43x because I am timing two batch variable deletions in these measurements, instead of just the constraint deletion, and they did not change performance with the changes introduced by this commit).

I executed the Gurobi tests as usual:

@odow can you give it a look? If you want me to stop pinging you in these cases just tell me, I am never sure if I should or not. If you think it needs extra tests or a "minimal working benchmark" I can work on them asap.

[odow]

You are welcome to ping me :)

Does this get called in the MOI tests? At minimum, we need to ensure that added code is called. This can either be in the MOI tests, or via a specific test in Gurobi.jl.

[henriquebecker91]

This is a screenshot of the Gurobi.jl tests after the last commit but with a println added in the code to be extra sure the method is called.

[henriquebecker91]

Thanks for the fast response!