nx_slice_pipe_v2_real_test.nx

nx_slice_pipe_v2_real_test.nx -- HARD REAL TEST of the integrated slicer on a non-trivial mesh (tetrahedron with real overhangs). Tetrahedron has 4 slanted faces -- overhang detection MUST fire on at least one layer pair, so the auto-supports pre-pass produces a non-empty plan. Industry slicers auto-detect supports too; ours auto-detects AND load-aware-sizes them in the same call. HARD: actual mesh geometry with real overhangs, not a synthetic cube/cube-with-perfect-edges that masks bugs. Real metric values asserted (byte count, layer count, support count) -- not just "PASS if it ran." Per the SUPERIOR-CAPABILITY cardinal: this is the outcome validation for the integrated slicer. Industry slicers can run on a tetrahedron too -- but they emit constant-diameter supports. Our test verifies the substrate produces the integrated artifact industry can't (auto-detect + load-aware-size). expect_exit: 0 license_tier: ORIGINAL