I'm reopening this after changing branches on my fork. The failing tests are expected as this PR will yield slightly different temperature interpolations (more accurate ones IMO). Let me know if I should update the tests.

This change is driven by the fact that, by default, the interp_linear_T function created by linear_interpolation throws when asked to extrapolate without being given an extrapolation method.

The need for extrapolation is due the 1d temperature profile being constructed by rounding min/max depth to the nearest km, sometimes resulting in a slightly smaller depth range than the actual layers in the model.

I think I understand the original impetus for rounding to the nearest km: it is highly likely that the user has input lithosphere thicknesses as a whole-km number, but finite grid spacing means that interfaces get shifted a bit.

However, I think it is more accurate to forego rounding here. Whether my lithosphere ends up being 100km or 99.3km, I want the base and surface of it to initially be certain temperatures in the model.

Removing the rounding removes the possibility of extrapolation and results in a T profile that will be interpolated onto the model slightly more accurately.

Also renamed some of the internal variables for clarity.