Aerosol activation has numerical instabilities

The ARG parameterization (`max_supersaturation`) is numerically unstable for certain atmospheric conditions.

For example:
* [Line 167](https://github.com/CliMA/CloudMicrophysics.jl/blob/6385546b3ba6f4e69f857c3a6e7b0f47e89ae6d4/src/AerosolActivation.jl#L167): `sqrt(α * w / G) `- when `α * w / G` becomes negative
* [Line 180](https://github.com/CliMA/CloudMicrophysics.jl/blob/6385546b3ba6f4e69f857c3a6e7b0f47e89ae6d4/src/AerosolActivation.jl#L182): `(ζ / η)^ap.p1` - when intermediate values become -Inf

These (and other issues) need numerical guardrails and tests.

Start in src/AerosolActivation.jl at the max_supersaturation implementation, especially lines 167 and 180. Examine how negative α * w / G and -Inf intermediate values arise under the atmospheric conditions described. Add numerical guardrails and tests covering these failure cases, then verify the aerosol activation behavior remains stable.

Written by the indexing model from the issue text.