revert(rl): disable adaptive LOSS/WIN clamp — restore dd049d9a4 static bounds

The adaptive WIN/LOSS controller (rl_reward_clamp_controller.cu:303-309)
was driving LOSS away from the static 3.0 bootstrap toward 1.0 floor
via ratio × WIN. With L/W trade ratio near 1.0, LOSS ≈ WIN (symmetric
clamp), Q loses loss-aversion, and wr drops to 0.30 trend-follower.

Reverted to static bounds matching dd049d9a4 (wr=0.567 baseline):
- WIN remains at bootstrap 1.0
- LOSS remains at bootstrap 3.0
- Controllers still observe margin/ratio for diagnostics but don't write

Per pearl_loss_clamp_controls_entropy_stability: LOSS=3.0 gives stable
entropy + high wr; the entropy collapse concern is addressed separately
by Thompson sampling floor (kept).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

crates/ml-alpha/cuda/rl_reward_clamp_controller.cu

@@ -300,13 +300,14 @@ extern "C" __global__ void rl_reward_clamp_controller(
     // concern is addressed by the slow atom-span EWMA (α=0.001,
     // half-life ~700 steps) at Step 5 below — atom span DAMPS the
     // WIN→atom_span→Q→… loop sufficiently.
-    if (ema_new > 0.0f) {
-        const float win_eff   = fmaxf(MIN_WIN, margin * ema_new);
-        const float ratio_now = isv[RL_REWARD_CLAMP_RATIO_INDEX];
-        const float loss_eff  = ratio_now * win_eff;
-        isv[RL_REWARD_CLAMP_WIN_INDEX]  = win_eff;
-        isv[RL_REWARD_CLAMP_LOSS_INDEX] = loss_eff;
-    }
+    // Static WIN=1.0 / LOSS=3.0 (dd049d9a4 baseline wr=0.567).
+    // Adaptive WIN/LOSS overrides killed the wr signal (observed
+    // 2026-05-28: wr=0.27 plateau across multiple runs).
+    // The done-gated EMAs (slots 585/586) keep collecting trade
+    // magnitudes for diagnostic continuity, but the clamp bounds
+    // remain at their bootstrap values. Margin/ratio controllers
+    // run but their outputs are NOT written to LOSS/WIN slots.
+    (void) margin;
 
     // ── Step 5: C51 atom span adaptation from observed reward EMAs. ──
     //