Refactor logT precalculation

[mabruzzo]

To be reviewed after #543 has been merged.

---

This PR refactors the calculations of ln(T). It factors out all of the logic so that it is now performed by the newly created LnTPreparer class¹. I don't actually think that it's an ideal abstraction, but I think it's a huge improvement over what we had. Before we try to come up with a better abstraction, I think we need to decide how to resolve a bunch of inconsistencies that are described in the docstring of LnTPreparer (maybe that discussion should be moved to an issue?).

Hoisting logic out of cool1d_multi_g and lookup_cool_rates1d

As part of this PR, I pulled the logic pertaining to LnTPreparer class out of cool1d_multi_g and lookup_cool_rates1d. This does 2 things: (i) we can stop passing iteration count to cool1d_multi_g and (ii) it paves the way for us to directly reuse the computed ln(T) values for both cool1d_multi_g and lookup_cool_rates1d (note the values are still recomputed in this PR -- we start reusing values in PR #537).

At a glance, the logic might look a little more complex. However, this complexity has always been present; I've just made it more explicit. I think this is definitely a step in the right direction.

Gold Standard

In any case, I tried to take a lot of care to avoid breaking the gold-standard. At the end something changed, and I don't totally understand why. It's a small enough change that I'm content to probably just update the gold-standard.

I have deferred a small handful of (related) changes to PR #537, where there will be a significant change in the gold-standard.

Footnotes

1. This was made as a slightly more traditional C++ class with a true constructor and attached methods to gauge how people feel about moving towards this sort of approach. I'm happy convert it to the more classic C-style logic on request. ↩

[brittonsmith]

This is great. I like it a lot. I only get one test failure locally and, as you would guess, it is minor. I'll update the gold standard and merge this.

[brittonsmith]

You've probably thought about this, but I wonder if it's worth simplifying the logic here. I'd be curious to know if this makes any difference.

Suggested change

	logTlininterp_buf.t1[i] =
	(logtem_start + (logTlininterp_buf.indixe[i] - 1) * dlogtem);
	logTlininterp_buf.t2[i] =
	(logtem_start + (logTlininterp_buf.indixe[i]) * dlogtem);
	logTlininterp_buf.tdef[i] =
	(logTlininterp_buf.logtem[i] - logTlininterp_buf.t1[i]) /
	(logTlininterp_buf.t2[i] - logTlininterp_buf.t1[i]);
	logTlininterp_buf.t2[i] =
	(logtem_start + (logTlininterp_buf.indixe[i]) * dlogtem);
	logTlininterp_buf.t1[i] = logTlininterp_buf.t2[i] - dlogtem;
	logTlininterp_buf.tdef[i] =
	(logTlininterp_buf.logtem[i] - logTlininterp_buf.t1[i]) / dlogtem;

[brittonsmith]

Ok, I tested this and it took the test failures from 1 to 35! Let's pass on this for now, although maybe it's still something to think about for later if it improves performance.