fix: Initialize unassigned total_soil_depth in soilLiqFlx.f90

[DarriEy]

Summary

• Initialize the local variable total_soil_depth in update_surfaceFlx_liquidFlux_computation_max_infiltration_rate (soilLiqFlx.f90)
• The variable is declared (line 865) but never assigned, causing undefined behavior
• Computes total_soil_depth = sum(in_surfaceFlx % mLayerDepth), consistent with how sum(mLayerDepth) is used inline in the master/develop branches

Problem

The Green-Ampt and TOPMODEL infiltration formulae divide by total_soil_depth (via depthWettingFront). When the variable is uninitialized:

• Linux x86-64 / gfortran 13: Stack reads as 0.0, causing depthWettingFront = 0.0 and 0/0 = NaN in the infiltration rate. The NaN propagates through the flux vector into computResid:

  FATAL ERROR: .../computResid/NaN in residuals
  Note: The following floating-point exceptions are signalling: IEEE_INVALID_FLAG
  

• macOS / ARM: The bug is masked because stack memory at that address happens to contain a non-zero value from a previous call frame.

Diagnosis

Debug prints added around the Green-Ampt computation confirmed:

DGA avCap= 0.798  rzLiq= 0.325  rzIce= 0.0
DGA dwf=   0.0    tsd=   0.0     ← total_soil_depth is uninitialized (zero)
DGA-GA Ks= 1.39E-06  wfs= 0.3   xMI= NaN

Building with -finit-real=snan confirms the variable is never assigned — the entire column 5 of the Jacobian becomes signaling NaN.

Test plan

• Verified fix on Ubuntu 24.04 / WSL2 / gfortran 13.3.0 with a 3-year SUMMA simulation (Bow at Banff, lumped basin, infRateMax=GreenAmpt, nrgConserv=enthalpyForm)
• Confirmed total_soil_depth value matches expected soil column depth (1.0 m for test domain)
• Simulation completes with "finished simulation successfully"

🤖 Generated with Claude Code

[ashleymedin]

Thanks Darri -- this is now fixed so we can close this without merging. @wknoben