nx_render_pass.nx

nx_render_pass.nx -- L4 mesh-render composer (vertex + zbuf + texture). Closes the GRAPHICS COMPOSITION gap. Today's substrate had: L1 containers : nx_image.Image + nx_image.ImageS64 L2 transforms : nx_raster (Bresenham + Pineda) + nx_zbuf (depth-tested triangle) + nx_tex_sample (UV bilinear + nearest) + nx_vertex_pipeline (world -> screen) L3 algorithms : the building blocks above L4 composer : MISSING This brick is the L4 composer. Caller passes a mesh (indices + vertices), a texture, an MVP matrix, and a framebuffer + depth buffer. We: 1. Run vertex_pipeline_transform to project world vertices to screen space. 2. For each triangle (i, j, k) in indices: pull the three transformed vertices, sample texture at their UVs, call nx_zbuf_triangle with the depth-tested color write. Pure i64 substrate. No FPU, no GPU, no driver. Runs on every backend including Cortex-M3 with a framebuffer. ===== Mesh format =============================================== verts: 4 i64 per vertex: (x_q14, y_q14, z_q14, packed_color) uvs: 2 i64 per vertex: (u_q10, v_q10) -- texture coords indices: 3 nx_int per triangle: (i0, i1, i2) -- mesh-flat layout All Q-format choices match nx_vertex_pipeline + nx_tex_sample conventions so the buffers compose directly. ===== Vertex coloring (v1) ====================================== v1 samples the texture at each VERTEX'S UV and lets nx_zbuf_triangle barycentric-interpolate the per-vertex sampled colors across pixels. This is the "Gouraud shading" approach -- fast, correct for diffuse Lambertian surfaces, but lacks per-pixel texture