Add thermocline with variable N²(z)

[ali-ramadhan]

This PR just changes the thermocline in three_layer_constant_fluxes.jl to have a variable N²(z) structure parameterized by a fitted cubic function.

The idea is to enrich the training data as convection is no longer entraining into a fluid layer of constant stratification, although might need to play around with the model parameters to get a temperature profile where the N²(z) structure is significant (and it still finishes in a reasonable amount of time).

To give some idea of what it looks like:

[ali-ramadhan]

Ah thankfully it was just a minor conflict.