Improve docs of `linfa-hierarchical`

[bytesnake]

The interface and documentation of linfa-hierarchical is not updated in a long time. This PR adds the example to the documentation and changes the implementation of transform to predict, which is more correct in this context.

Currently blocked on a cyclic dependency caused by implementing linfa::Predict for linfa::PredictRef with kernel data. A workaround is by adding a feature flag to linfa-kernel which removes all linfa related parts and including that into the dependency tree for linfa.

[YuhanLiin]

Why do we need to reverse the dependency between linfa and linfa-kernel?

[bytesnake]

to use the Kernel structure in the implementation of Predict for T where T: PredictRef