aerocore: defects in the aerosol path

[Sohex]

The aerosol module compiles and is reachable through l_aero, and carries several independent defects that would surface on a first use of that path. They are fixed together because they are found together and none is usefully separable.

Offered as groundwork rather than as a fix to observed behaviour. With l_aero at its default the module does nothing, so no current run changes; what changes is that the path is correct for anyone who turns it on.

The diff is commented per defect. Compiles cleanly.

I am happy to split this into one commit per defect if you would rather review them separately.

[Sohex]

A cross-check that may help when you and Maureen scope the re-run: the viscosity exponent in this PR is confirmed by Cohen et al. (2024)'s own appendix.

Their eq. A5 gives the Rosner (1986) parameterisation as

η = 5·√(π·m·k_B·T)·(k_B·T/ε)^0.16 / (16 · 1.22 · π·d²)

with m = 4.652e-26 kg, d = 3.64e-10 m and ε/k_B = 95.5 K for N₂. The code writes that exponent as (4/25) — which in Fortran is integer division and evaluates to 0, so (temp/eps)**0 = 1 and the temperature dependence is deleted outright. Every other factor in the expression matches A5 exactly; it is only the exponent that is lost.

So the paper documents 0.16 and the model computed 0. The size of it, using their own ε:

T (K)	(T/95.5)^0.16	viscosity low by	Stokes settling fast by
200	1.126	11.2%	12.6%
250	1.167	14.3%	16.6%
300	1.201	16.7%	20.1%

Settling goes as 1/η, so for temperate rocky planets around 250 K the particles were falling roughly 17% too fast. That is a systematic bias in the settling scheme rather than noise, and it is in the same direction everywhere, so it should show up as a coherent shift in the equilibrium haze burden rather than as scatter.

Worth pairing with #60 (the radiation was using radmod's hardcoded 50 nm rather than the configured radius) and #61 (the bottom-layer sink) when deciding what a re-run needs to carry.