Delegate meteoSI.py to the standalone meteo_si package

[maahn]

Delegate meteoSI.py to the standalone meteo_si package

python/pyPamtra/meteoSI.py duplicated a lot of general-purpose humidity/
density/temperature formulas that already exist, independently maintained
and tested, in maahn/meteo_si (same author, MIT-licensed, on PyPI). This
switches the functions that already match to thin delegating wrappers
(e2q, q2e, T_virt_q, T_virt_rh, moist_rho_q, moist_rho_rh, rh2q, rh2a,
a2rh, q2rh, rh_to_iwv, usStandard), re-exports the shared constants from
meteo_si.constants, and keeps only the genuinely PAMTRA-specific
adiabatic-LWC cloud physics (detect_liq_cloud, adiab, mod_ad,
pseudoAdiabLapseRate, vaphet) implemented locally, since meteo_si has no
equivalent for those.

One finding shaped the whole refactor: meteoSI.e_sat_gg_water (Goff &
Gratch 1946) and meteo_si.humidity.e_sat_gg_water (WMO CIMO Guide 2008)
are two different formulas that happen to share a name -- confirmed by
diffing both sources, not by name alone. src/e_sat_gg_water.f90 (the
Fortran core) uses the same Goff-Gratch formula as the Python side, and
the two need to stay numerically consistent since Python builds the
profiles the Fortran engine then runs radiative transfer on. So this
isn't a plain find-and-replace: meteo_si gained a new, distinctly-named
e_sat_goffgratch_water (this repo's own PR to that project, merged and
released as meteo_si 0.2.0), and every function here that depends on
saturation vapor pressure (rh2q, rh2a, a2rh, q2rh, moist_rho_rh,
rh_to_iwv) passes that in explicitly via meteo_si's existing e_sat_func
override parameter. PAMTRA's own e_sat_gg_water name is preserved as a
one-line alias to it, so nothing calling pyPamtra.meteoSI.e_sat_gg_water
(including tests/test_meteosi.py) needs to change.

Incidental fix: the old rh_to_iwv was actually broken (referenced the
undefined name meteoSI.rh2q/meteoSI.moist_rho_q from inside the same
module, and a bare exp that doesn't exist -- would NameError if ever
called). It's dead code today (grepped -- nothing in this codebase calls
it), but delegating to meteo_si's working version fixes it anyway.

usStandard (1976 US Standard Atmosphere) moved to meteo_si too, as a new
atmosphere.py module there (also part of the same PR/release) -- the
underlying algorithm is Ralph Carmichael's public-domain Fortran (PDAS),
and this was Maahn's own Python translation of it, previously living only
here.

pyproject.toml/pixi.toml: added meteo_si>=0.2 as a dependency (pixi via
[pypi-dependencies], since meteo_si isn't on conda-forge yet -- no
conda-forge/meteo_si-feedstock exists -- pixi mixes conda + PyPI deps
natively). conda-recipe/recipe.yaml also lists it in requirements.run
(its --no-deps pip install needs every dependency spelled out there) with
a comment on the same conda-forge gap; that recipe isn't usable for an
actual conda-forge build until meteo_si has a feedstock, which is a
separate submission process out of scope here.

Verified: full pytest suite (66/66), specifically test_meteosi.py and
test_regression.py (golden-output radiative-transfer regression) --
proof that routing humidity/density conversions through meteo_si with
the Goff-Gratch override didn't change a single computed value.

Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com