Audit cache friendliness of Cholesky rrule

In #44, I directly ported over code from Nabla, which contains the following comment:

See [1] for implementation details: pages 5-9 in particular. The derivations presented in
[1] assume column-major layout, whereas Julia primarily uses row-major. We therefore
implement both the derivations in [1] and their transpose, which is more appropriate to
Julia.

[1] - "Differentiation of the Cholesky decomposition", Murray 2016

Julia actually uses column-major layout for its arrays. I haven't dug into the code to determine whether the implementation matches the comment, i.e. assumes row-major, but it would be worthwhile to do so to ensure we're not being unnecessarily inefficient.