Fix inverted HYDCHR scaling for partially penetrated HFB layers

[bdestombe]

Problem

get_hfb_spd scales hydchr by the penetrated fraction of the layer that contains the bottom of the barrier:

hydchr_frac = (depth - layer_top_depth) / thicki[ilay]
spd.append([cellid1, cellid2, hydchr * hydchr_frac])

But a lower HYDCHR is a stronger barrier in MF6 (the barrier conductance is HYDCHR × face area, placed in series with the cell-to-cell conductance). So a wall that penetrates only 10% of a layer is modeled as a barrier ~10× tighter than a fully penetrating wall, instead of leaving the un-penetrated 90% of that face open. The effect is upside-down: the less a barrier penetrates a layer, the more it blocks it.

A two-cell MF6 experiment (10 m layer, K = 10 m/d, unit gradient):

face	flow
no barrier	100.0 m³/d
full wall, hydchr = 0.01	9.09 m³/d
nlmod scaling for 10% penetration (hydchr = 0.001)	0.99 m³/d
physical target for 10% penetration	≳ 90 m³/d

Fix

For the layer containing the barrier bottom, use a parallel-path equivalent characteristic: the open strip over the un-penetrated fraction 1 − f in parallel with the walled strip over f (the latter in series with the aquifer). Equating the MF6 series form to that combination gives

hydchr_eff = (hydchr + (1 − f) · c) / f

with c = kh_harmonic / distance the open-face conductance per unit area. This reduces to hydchr at full penetration (f = 1) and leaves the face open (hydchr_eff → ∞) as f → 0.

Validation

New test test_get_hfb_spd_partial_penetration_matches_resolved_barrier compares the equivalent single-layer barrier against a reference MF6 run that resolves the partially penetrated layer into a walled and an open sublayer — the flows match within 1%. The previous scaling underestimates that reference flow by more than a factor of 10. The existing structured/vertex/multilinestring expectations were updated to the parallel-equivalent values. Full tests/test_023_hfb.py passes (23 tests).

Surfaced while reviewing the Bergen sheet-pile walls in the NHFLO 09pwnmodel2 model, which drive hfb_from_df → get_hfb_spd with depth barriers.

[dbrakenhoff]

Good fix! As a side note, I think modflow also accepts negative hydchr in which case it is interpreted as a factor on the cell-to-cell conductance, which could be useful in the context of a partially penetrating impermeable walls. Not sure if we want to deal with that here, but just mentioning it in case it is of value.