fix: close round-2 review findings — Critical silent-note bug plus majors/minors

Fixes the 2-point spline-EG early-free bug causing silent fade-ins, the pitch/filter enable-toggle Trigger-forcing hole, the contour-node/marker pixel shadow, missing deck-residue test coverage, and stale comments in knob_deck and spline_edit.

src/core/instrument/engine/velocity_curve.h

@@ -229,11 +229,16 @@ public:
     }
     void clear() { pts_ = nullptr; n_ = 0; }
     bool active() const { return n_ >= 2; }
-    // True once the cursor has settled on the contour's LAST segment: past this point there is
-    // no further point to rise into, so a value read here that reaches 0 is a genuine permanent
-    // terminus (Voice::tickAmplitude's early-free), unlike a 0 touched mid-contour, which a
-    // later segment may still rise out of (the spline is deliberately not globally monotone).
+    // True once the cursor has settled on the contour's LAST segment. On its own this does NOT
+    // make a 0 read here a terminus: the final segment's LEFT endpoint can also be 0 (a 2-point
+    // contour is nothing but a single "final" segment starting at frame 0), which would read 0
+    // while about to rise. Voice::tickAmplitude pairs this with terminalValue() == 0 — the
+    // segment's RIGHT endpoint, i.e. the whole contour's true end — before calling a 0 read the
+    // note's genuine permanent terminus.
     bool onFinalSegment() const { return seg_ + 2 == n_; }
+    // The current segment's right endpoint — the whole contour's terminal Y only when paired
+    // with onFinalSegment() (see there).
+    double terminalValue() const { return y1_; }
 
     // `phase` is normalized position over the contour's whole span, [0,1]; out-of-range clamps
     // to the terminal values (a note past its span holds the contour's last level).

src/core/instrument/engine/voice.h

@@ -186,12 +186,18 @@ private:
         if (ampSplineCur_.active() && playMode_ == PlayMode::Trigger) {
             // A contour covers the sample end to end, so the head leaving the span IS the end of
             // the note — the exhaustion path in advanceFrame is what frees the voice. A contour
-            // that flatlines at 0 across its FINAL segment is a permanent terminus (no later
-            // segment to rise out of), so that case frees early too, the spline analogue of a
-            // staged AHD's finished() — mid-contour dips do not, since the spline is deliberately
+            // whose TERMINAL value (the final segment's right endpoint) is 0 reaches a genuine
+            // permanent terminus early, the spline analogue of a staged AHD's finished(). Gating
+            // on the terminal value, not just the segment index, matters because a 2-point
+            // contour IS a single "final" segment from frame 0 — checking onFinalSegment() alone
+            // would call a contour that STARTS at 0 (e.g. a fade-in) over before it ever rises.
+            // Mid-contour dips through 0 still don't free early, since the spline is deliberately
             // not globally monotone.
             amp = ampSplineCur_.eval(splinePhase());
-            if (amp == 0.0 && ampSplineCur_.onFinalSegment()) amplitudeDone_ = true;
+            if (amp == 0.0 && ampSplineCur_.onFinalSegment() &&
+                ampSplineCur_.terminalValue() == 0.0) {
+                amplitudeDone_ = true;
+            }
         } else if (playMode_ == PlayMode::Gate) {
             amp = env_.tick();
             if (env_.finished()) amplitudeDone_ = true;