Snapshot: fog implementation + fidelity tooling baseline (pre bilinear-clamp fix)

Per-vertex GS fog end-to-end (gs_stub emit incl. persp_emit5, gs_prim_list_feeder
XYZ2->XYZF2 on PRIM.FGE, gs_make_sh3_scheduler_fixture.py F/FGE packing), new fog
TBs, fidelity attribution tooling. Functional baseline before removing the dead
bilinear lerp8 clamps (Codex: 161-node comb loop -> -0.042ns setup fail).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tools/preview_scanout_ch438.py

@@ -0,0 +1,74 @@
+#!/usr/bin/env python3
+"""Preview the exact Ch438 source-space 3x3 binomial scanout."""
+import math
+import sys
+
+from PIL import Image
+
+
+def binom3(a, b, c):
+    return tuple((x + 2 * y + z + 2) // 4 for x, y, z in zip(a, b, c))
+
+
+def score(a, b):
+    ae = 0
+    se = 0
+    n = a.width * a.height * 3
+    for pa, pb in zip(a.getdata(), b.getdata()):
+        for ca, cb in zip(pa, pb):
+            d = int(ca) - int(cb)
+            ae += abs(d)
+            se += d * d
+    return ae / n, math.sqrt(se / n)
+
+
+def main(argv):
+    if len(argv) != 4:
+        raise SystemExit(
+            "usage: preview_scanout_ch438.py <fb.png> <reference.png> <out.png>"
+        )
+
+    src = Image.open(argv[1]).convert("RGB")
+    if src.size != (640, 480):
+        raise SystemExit(f"expected 640x480 framebuffer, got {src.size}")
+    ref = Image.open(argv[2]).convert("RGB").resize(
+        (640, 480), Image.Resampling.BILINEAR
+    )
+
+    # RTL rounds after each separable pass, so mirror that order exactly.
+    horizontal = Image.new("RGB", src.size)
+    hp = horizontal.load()
+    for y in range(480):
+        for x in range(512):
+            xm1 = max(0, x - 1)
+            xp1 = min(511, x + 1)
+            hp[x, y] = binom3(src.getpixel((xm1, y)),
+                              src.getpixel((x, y)),
+                              src.getpixel((xp1, y)))
+
+    filtered = Image.new("RGB", src.size)
+    fp = filtered.load()
+    for y in range(32, 480):
+        ym1 = max(32, y - 1)
+        yp1 = min(479, y + 1)
+        for x in range(512):
+            fp[x, y] = binom3(horizontal.getpixel((x, ym1)),
+                              horizontal.getpixel((x, y)),
+                              horizontal.getpixel((x, yp1)))
+
+    out = Image.new("RGB", (640, 480))
+    px = out.load()
+    for y in range(480):
+        sy = 32 + (y * 14) // 15
+        for x in range(640):
+            sx = (x * 4) // 5
+            px[x, y] = filtered.getpixel((sx, sy))
+
+    out.save(argv[3])
+    mae, rmse = score(out, ref)
+    print(f"binomial3x3 MAE={mae:.4f} RMSE={rmse:.4f}")
+    print(f"wrote {argv[3]}")
+
+
+if __name__ == "__main__":
+    main(sys.argv)