Fix cvmix_KPP ice scaling: apply (1-Aice) to ustar and Stokes wind velocity

[patrickscholz]

Summary

Affects only the cvmix_KPP mixing scheme (kpp_reduce_tauuice = .true.). All other mixing schemes (PP, TKE, etc.) are unaffected.

When kpp_reduce_tauuice = .true., the friction velocity ustar was scaled by (1-Aice)^2 to suppress wind-driven boundary layer turbulence under sea ice, but the wind speed inputs to the Stokes drift / Langmuir turbulence parameterisation (wind_norm, u_wind, v_wind) were left at their full open-ocean values. This created two problems:

1. EFactor blow-up under high ice concentration
The Langmuir enhancement factor is EFactor = sqrt(1 + c/LaSL^4) where LaSL = sqrt(ustar / uv_SLmean) and uv_SLmean ∝ 0.0162 * wind_norm. As a_ice → 1, ustar → 0 while uv_SLmean remained finite, driving LaSL → 0 and EFactor → ∞ — spurious infinite Langmuir turbulence enhancement directly under sea ice.

2. Unphysical Stokes drift under ice
The Stokes drift parameterisation uses a Pierson-Moskowitz fully-developed sea approximation, which assumes waves are locally generated by the local wind. Sea ice suppresses local wave growth, so applying the full open-ocean wind speed in ice-covered cells is physically inconsistent.

Changes (src/cvmix_driver/gen_modules_cvmix_kpp.F90)

• ustar scaling corrected: (1-Aice)^2 → (1-Aice). Physical basis: effective wind speed at the ocean surface is reduced by (1-Aice) due to ice sheltering → tau ∝ (1-Aice)^2 → ustar = sqrt(tau/rho) ∝ (1-Aice). The original ^2 line is retained as a comment (it was ported from MPIOM without prior testing in FESOM).
• New u_wind_eff / v_wind_eff variables: scaled by (1-Aice) when kpp_reduce_tauuice = .true., otherwise equal to u_wind / v_wind. Using the same factor as ustar preserves LaSL = sqrt(ustar/us) — both numerator and denominator scale identically, keeping EFactor finite.
• All Stokes wind inputs updated:
  • wind_norm (feeds cvmix_kpp_ustokes_SL_model and compute_Efactor, called twice)
  • Three calls to compute_stokes_velocities_MOM6style (two in the StokesMOST do nz loop, one after OBL depth is known)

Notes

• No behavioural change when kpp_reduce_tauuice = .false. (default); u_wind_eff = u_wind exactly.
• The TKE Langmuir scheme (tke_dolangmuir) is unaffected — it derives its Stokes proxy from the wind stress, which is inherently reduced under ice.
• For reference: MOM6 handles this differently, suppressing the Langmuir enhancement multiplier (lamult = 1) above a 5% ice threshold at the coupler level rather than scaling wind inputs inside KPP.

[JanStreffing]

Verified the bug and fix end-to-end. Should we have a test run of this?

[patrickscholz]

Xuejing Chen found that i think she will make some test, i dont think anybody else except her is using this in moment. Also our default option for kpp_reduce_tauuice=.false. so far!