nx_procgen_preset_test.nx

nx_procgen_preset_test.nx -- smoke for bidirectional progressive disclosure. What this validates -- the STRUCTURAL invariants of the cardinal: 1. Tier 0 just-works (single call, no knobs, returns useful struct) 2. Tier 0 == Tier 1 with defaults (no hidden code path under Tier 0) 3. Determinism: same (seed, preset, dims) -> bit-identical heightmap 4. Presets actually affect output (different preset -> different heightmap) 5. Tier-2 raw composition reproduces Tier-1 result (visible recipe) 6. Defensive at boundaries: invalid preset / zero dims fall back cleanly 7. Sealed-enum preset validity gate 8. Height-band classifier composes with Perlin sealed-enum 9. Poisson density honors preset radius differences (after the r>=10 saturation fix landed -- previously this assertion blocked on a substrate compiler-bug masquerading as algorithm bug; see project-poisson-disk-bug-large-r-2026-05-15.md).