Generalise the length scale to multiple dimensions in Kernel class

Generalise `length_scale` to multiple dimensions in the Kernel class.

This will allow for anisotropic kernels, where each dimension of the length scale defines the length-scale of the respective feature dimension. This is useful when different features have different levels of relevance or influence on the output.

Start by locating the Kernel class and reading how length_scale is currently represented and applied. Generalize it so each feature dimension can use its corresponding scale for anisotropic kernels, then verify that existing scalar behavior remains supported and that per-dimension scaling works as intended.

Written by the indexing model from the issue text.