nx_perceptual_render_test.nx

nx_perceptual_render_test.nx -- G1+ end-to-end per-perceptual-profile rendering KAT. Closes the cardinal structural diff vs DirectX with viewable artifacts: the SAME triangle, rendered at the SAME coordinates, produces DIFFERENT PPM byte sequences depending on the declared nx_perceptual_profile. No other graphics stack on earth does this; every incumbent bakes in human-trichromat sRGB defaults. Pipeline per profile: pack red (255,0,0,255) via nx_fb_pack_for_profile(profile) -> profile-specific i64 packed cell value -> nx_raster_triangle paints triangle interior pixels with that cell -> nx_swap_present serializes fb -> PPM bytes -> KAT verifies the data area's RGB bytes at a known interior pixel match the per-profile expectation. Three profiles exercised end-to-end: HUMAN_NORMAL : RGBA8888 (R, G, B, A in bytes 0..3) pack(255,0,0,255) -> 0xFF0000FF -> PPM RGB (255, 0, 0) DOG_VIZSLA : dichromat Y+B in bytes 0..1 pack(255,0,0,255) -> 0x0000004D -> PPM RGB (77, 0, 0) Y = (77*255 + 150*0 + 29*0 + 128) / 256 = 77 HONEYBEE : G+B+UV+A in bytes 0..3 (no R sensitivity) pack(255,0,0,255) -> 0xFF000000 -> PPM RGB (0, 0, 0) (red invisible to honeybee perceptual profile; PPM extracts low 3 bytes; UV in byte 3 not visible in 3-channel PPM -- 4-channel PAM at G2) AVIAN tetrachromat is intentionally NOT in this demo: its packer puts R in byte 0, G in byte 1, B in byte 2, UV in byte 3, so the PPM RGB extraction sees identical bytes to HUMAN_NORMAL. Showing the AVIAN UV-channel delta needs PAM or a custom UV-PPM extension, queued for G2. This is honestly noted -- not a hack, just a 3-channel PPM limit. Composes: nx_framebuffer.nx (fb alloc + clear + cell layout) nx_fb_perceptual.nx (per-profile packers; G1 shipped) nx_raster_sw.nx (Pineda triangle rasterizer; G1 shipped)