remove scalar->vector gradients

[ArnoStrouwen]

This is a breaking change.
Should the function be kept around but call jacobian under the hood, or do we remove it entirely?
Upstream tgrad depends on this, so needs to be nicely deprecated.

[ChrisRackauckas]

Should we fold this into derivative!?

[ArnoStrouwen]

I was thinking the same, because that is what ForwardDiff does,
but then the issue is that the derivative of a vector already has the meaning of taking the derivative of a scalar -> scalar function at each point in the vector, so adding scalar->vector functions will mess up the dispatch.
Maybe rename that multipoint functionality to FiniteDiff.finite_difference_derivatives with an s?

[ChrisRackauckas]

Sounds fine to me.