Issues with requesting grad for predefined kernels

[Referee1405]

There appear to be multiple issues if grad is set to "true" for the predefined kernels (Laplace and Helmholtz at least). Interestingly enough, there is a version that works:

using IFGF
targets = rand(3,10)
sources = rand(3,9)
charges = rand(9)
dipvecs = rand(3,9)
IFGF.laplace3d(targets, sources, dipvecs=dipvecs, grad=true, tol=1e-5)

But if charges is set to a vector, an error occurs. The error also changes depending on if dipvecs is set or not. The main examples that cause issues are

IFGF.laplace3d(targets, sources, charges=charges, grad=true, tol=1e-5)
IFGF.laplace3d(targets, sources, charges=charges, dipvecs=dipvecs, grad=true, tol=1e-5)

Similar calls to Helmholtz also yields these issues.