Account for coat IOR in base roughening

[portsmouth]

As discussed on Slack, the coat roughening formula we suggest in the spec:

incorrectly ignores the IOR of the coat. For example if the coat IOR is 1, the coat is effectively not present so there should be no roughening effect.

Generally, I think we need to do some simulations to check the actual behavior, and come up with a good fitting formula, as the theory seems a bit hand wavy at present.. At least it should be understood that these formulas are just suggested approximations, but it's not good if they're totally wrong in certain regimes.

[portsmouth]

Peter's suggestion on Slack for the IOR dependence (based on some hand-wavy but reasonable arguments) was:

r_b^\prime = \left(  r_b^4 + X r_c^4  \right)^\frac{1}{4}

where

X = \begin{cases}
			\mathrm{lerp}(1, 0, \eta^{-1}_c), & \text{if  } \eta_c \ge 1\\
                         \mathrm{lerp}(1, 0, \eta_c), & \text{otherwise}
		 \end{cases} \ .

This seems fine to me as the suggestion we put in the spec. Obviously, this is essentially a guess which has the right qualitative behavior, nothing more. Some more quantitative evaluation of the accuracy is planned.

I suggest a slightly more compact formula which does the same as the above, though:

X = 1 - \min(\eta_c, \eta_c^{-1}) \ .

[portsmouth]

Also we shouldn't be explicit about the lerp formula with the coat weight, in (61) above. As physically if the coat is partially present, the result will a blend of the original and roughened base lobes, which doesn't look the same as a single lobe roughened by the mixed roughness. Or at least, point out that the use of the mixed roughness is a further approximation.