Describe the workflow you want to enable

Currently KernelDensity.sample only works for the gaussian and tophat kernels, but I'd like to add support for the linear and exponential kernels as well.

Describe your proposed solution

The sampling process amounts to picking a point from the original dataset at random, then sampling from the kernel distribution and adding it to that point. The laplace distribution from numpy can be used for the exponential kernel and the triangular distribution from numpy can be used to generate samples from the linear kernel, so this change is only adding a few extra lines, plus some tests. I have a working version for this so will put in a PR.

Adding for linear and exponential for now as they are the easiest to implement. epanechnikov and cosine are possible as well, but are slightly more involved.

Describe alternatives you've considered, if relevant

No response

Additional context

No response