fix(scoring): correct CBF baseRate denominator — Jøsang SL §12.3

aDenomBase must be u_A + u_B - 2·u_A·u_B.
Previous code used (denom - 2·u_A·u_B) where denom already included
-u_A·u_B, making it u_A + u_B - 3·u_A·u_B — off by one term.

With equal baseRates (always 0.5) the correct formula collapses to 0.5.
The wrong formula produced baseRate ≈ 1.9 when uncertainty is high,
causing projectScore() to exceed 1.0 (→ 100+ clipped to 100).

Visible symptom: github:tankcdr scored 100 with weak signals (github
score 0.50, twitter score 0.20) while github:vbuterin scored 78 with
strong signals (github score 0.80, confidence 0.95). After fix:
tankcdr ≈ 46, vbuterin ≈ 79.

Author: tankcdr

packages/core/src/engine/scoring.ts

@@ -51,8 +51,12 @@ export function fuseTwo(a: Opinion, b: Opinion): Opinion {
   if (Math.abs(a.uncertainty - 1) < 1e-10) return { ...b };
   if (Math.abs(b.uncertainty - 1) < 1e-10) return { ...a };
 
+  // Correct per Jøsang SL §12.3: denominator for baseRate is u_A + u_B - 2·u_A·u_B
+  // NOT (u_A + u_B - u_A·u_B) - 2·u_A·u_B which was off by one u_A·u_B term.
+  // When both baseRates equal 0.5 (always) this correctly collapses to 0.5;
+  // the wrong formula produced baseRate ≈ 1.9 when uncertainty is high, inflating scores past 100.
   const aDenomBase =
-    denom - 2 * a.uncertainty * b.uncertainty;
+    a.uncertainty + b.uncertainty - 2 * a.uncertainty * b.uncertainty;
 
   return {
     belief: