Internal function `neg_energy` should be called `neg_kinetic_energy`

[nsiccha]

See e.g. https://github.com/TuringLang/AdvancedHMC.jl/pull/439/files#r2549647644

[THargreaves]

This is related to the confusing use of the DualValues in phasepoint [1].

I think most people would assume each DualValue would have the Hamiltonian energy as the value term with different grads for θ and r (that's what I did at first), but the Hamiltonian is actually split up across the two terms. This makes sense computationally but is a bit confusing with clarification.

[1] https://github.com/TuringLang/AdvancedHMC.jl/pull/439/changes#diff-492cca51aebd2b83e9b13106eccdf610b5d4fb7ecee74dd2066b3573f685d06cR110-R111

[nsiccha]

I agree, see e.g. also #458 (comment) which relates to #173, which I'd close in favour of dealing with these quantities "properly".