cpgrid_from_horizons: zero-thickness pinched cells survive as active with NaN centroids

[tomerarnon]

Summary

When cpgrid_from_horizons generates a grid with zero-thickness cells (from horizons that are equal at some pillar columns), the pinched cells survive into the final mesh as active cells with zero volume and NaN centroids. mesh_from_grid_section does not deactivate them because programmatically generated grids lack PINCH keyword data — the pinch processing path relies on Eclipse deck metadata that doesn't exist here.

Minimal reproducer

using GeoEnergyIO: cpgrid_from_horizons, mesh_from_grid_section
using JutulDarcy: reservoir_domain

# 3×3 grid, 2 layers. Layer 2 is pinched to zero thickness at column (2,2).
X = range(0.0, 300.0; length=4)
Y = range(0.0, 300.0; length=4)

h1 = fill(100.0, 4, 4)   # top
h2 = fill(200.0, 4, 4)   # middle
h3 = fill(300.0, 4, 4)   # bottom

# Pinch: make h2 == h3 at pillars surrounding column (2,2)
for j in 2:3, i in 2:3
    h2[i, j] = h3[i, j]
end

data = cpgrid_from_horizons(X, Y, [h1, h2, h3])
mesh = mesh_from_grid_section(data)
domain = reservoir_domain(mesh)

cc = domain[:cell_centroids]
vols = domain[:volumes]
nc = size(cc, 2)

println("Total cells: $nc (expected 17, got 18)")  # pinched cell not deactivated
nan_count = count(i -> any(isnan, @view cc[:, i]), 1:nc)
zero_vol = count(==(0.0), vols)
println("NaN centroids: $nan_count, zero-volume: $zero_vol")
# Output:
#   Total cells: 18 (expected 17, got 18)
#   NaN centroids: 1, zero-volume: 1

Expected behavior

The pinched cell should be excluded from the mesh (deactivated via ACTNUM or equivalent), producing 17 active cells with no NaN centroids and no zero-volume cells.

Environment

• GeoEnergyIO v1.1.31
• Julia 1.11.8

[tomerarnon]

Cross-section illustrating the issue: erosion pushes horizon 2 (blue) down to meet horizon 3 (bottom) at the center column, creating a zero-thickness cell (red). cpgrid_from_horizons generates this cell, but it survives mesh construction as an active cell with vol=0 and centroid=NaN.

[moyner]

Thanks for reporting! Can you try this branch and see if it helps? #45

[tomerarnon]

Yes, #45 seems to fix it!