More intuitive interface for batch computation

[shinaoka]

Based on feedback from my students, the current interface based on inheritance is not intuitive.
How about adding a new keyword parameter batchedfunc to optimize! and crossinterpolate and so on?
This is similar to scipy optimizer interface, where we can pass a function computing derivative, with jac.

@rittermarc
What do you think?

[rittermarc]

Is your proposal to just add a boolean batchedfunc, in which case the function will receive sets of indices instead of being evaluated on each element?

I agree that the current interface is not very intuitive, but I'm not sure how much of an improvement this would be.

[shinaoka]

If your function f object supports both:

crossinterpolate2(Float64, f, ....; batchedfunc=f, ....)

If you have two distinct functions:

crossinterpolate2(Float64, f, ....; batchedfunc=f2, ....)