Automatic differentiation in chunks

[jdeldre]

This PR enables carrying out automatic differentiation of functions with respect to positions and strength of elements in chunks that are smaller than the vector of elements. This should make the differentiation (gradient, jacobian) faster for larger systems. Consider, e.g.,

nblob = 15
pos = rand(ComplexF64,nblob)
str = rand(length(pos))
σ = 1e-2
blobs = Vortex.Blob.(pos,str,σ)

and we wish to differentiate velocity evaluated at a single point,

z = rand(ComplexF64)
compute_velocity(v) = induce_velocity(z,v,0.0)

with respect to blob positions. To enable the differentiation in chunks of size 3, for example, set

N = 3
cfg = PotentialFlow.Utils.ComplexGradientConfig(compute_velocity,blobs,PotentialFlow.Utils.Chunk{N}())
dwdz, dwdzstar = PotentialFlow.Elements.gradient_position(compute_velocity,blobs,cfg)