Problems linked to parallelism in aeroptproperties

[emillour]

When testing the Generic PCM in "earth-like" configuration, results obtained in serial or MPI differ (1-core run does give the same result as the serial run). Preliminary investigations point to the issue being in computed aerosol properties which differ with number of MPI cores...

After some investigations, it turns out that the poblem comes from the fact that aeroptproperties is designed to only handle aerosols with a fixed (over the whole planet and in time) effective variance whereas h2o_reffrad
returns a variance which varies (between 0.09 and 0.13) with location and time.

To fix this we revert to imposing a fixed variance in h2o_reffrad, as implemented in r2957