feat(apple): iPad ⌃⌥⇧Q release chord + click-to-recapture, pixel-grid snap, match-window opt-in

- InputCapture / StreamViewIOS: iPad ⌃⌥⇧Q un-capture chord, recognized from the
  GCKeyboard HID stream (no NSEvent monitor on iOS) for cross-client parity with
  the macOS/Windows/Linux combo; and a click into the video re-engages capture —
  the iPad analogue of macOS mouseDown → engageCapture(fromClick:), with the
  engaging click suppressed toward the host.
- SessionPresenter: snap the aspect-fit sublayer frame to the backing pixel grid.
  AVMakeRect centers the fit rect at fractional points, so the compositor
  resampled the layer — a uniform "everything soft" blur even when the drawable
  was pixel-exact 1:1. Rounding origin + size to device pixels makes the composite
  a true 1:1 blit; idempotent when already aligned.
- MetalVideoPresenter: PUNKTFUNK_BILINEAR_LUMA=1 A/B lever — compiles the shader
  with Catmull-Rom luma off (plain bilinear) to isolate bicubic overshoot from
  upstream fringing.
- SettingsView / StreamView / StreamViewIOS: match-window reverted to opt-in
  (default OFF) — the explicit mode is used and never auto-resized unless enabled.

clients/apple/Sources/PunktfunkKit/Video/MetalVideoPresenter.swift

@@ -124,7 +124,16 @@ float2 chromaUV(texture2d<float> lumaTex, texture2d<float> chromaTex, float2 uv)
 float3 sampleRgb(texture2d<float> lumaTex, texture2d<float> chromaTex, float2 uv,
                  constant CscUniform& csc) {
     constexpr sampler s(filter::linear, address::clamp_to_edge);
-    float3 yuv = float3(catmullRomLuma(lumaTex, s, uv),
+#ifdef PF_BILINEAR_LUMA
+    // DEBUG (PUNKTFUNK_BILINEAR_LUMA=1): plain bilinear luma — Catmull-Rom OFF. A/B lever to see if
+    // the bicubic overshoot contributes to edge fringing. NOTE: at a true 1:1 present both paths
+    // reduce to the identity texel, so if this toggle VISIBLY changes the picture, the present is
+    // NOT 1:1 (there's a resample); if it looks identical, the fringing is upstream (codec/source/OS).
+    float lumaY = lumaTex.sample(s, uv).r;
+#else
+    float lumaY = catmullRomLuma(lumaTex, s, uv);
+#endif
+    float3 yuv = float3(lumaY,
                         chromaTex.sample(s, chromaUV(lumaTex, chromaTex, uv)).rg);
     return saturate(float3(dot(csc.r0.xyz, yuv) + csc.r0.w,
                            dot(csc.r1.xyz, yuv) + csc.r1.w,
@@ -250,7 +259,16 @@ public final class MetalVideoPresenter {
         let pipelineHDR: MTLRenderPipelineState
         let pipelineHDRToneMap: MTLRenderPipelineState?
         do {
-            let library = try device.makeLibrary(source: shaderSource, options: nil)
+            // DEBUG A/B lever: PUNKTFUNK_BILINEAR_LUMA=1 compiles the shader with Catmull-Rom OFF
+            // (plain bilinear luma) by prepending a #define ahead of the source. Default (unset) is
+            // the normal bicubic path. Read at presenter creation — set it in the environment and
+            // relaunch to flip; the log line confirms which path built.
+            let bilinearLuma = ProcessInfo.processInfo.environment["PUNKTFUNK_BILINEAR_LUMA"] == "1"
+            let source = (bilinearLuma ? "#define PF_BILINEAR_LUMA 1\n" : "") + shaderSource
+            if bilinearLuma {
+                presenterLog.info("stage2: PUNKTFUNK_BILINEAR_LUMA=1 — Catmull-Rom luma DISABLED (bilinear)")
+            }
+            let library = try device.makeLibrary(source: source, options: nil)
             let vtx = library.makeFunction(name: "pf_vtx")
             let sdr = MTLRenderPipelineDescriptor()
             sdr.vertexFunction = vtx
@@ -590,8 +608,17 @@ public final class MetalVideoPresenter {
         let sig = "\(Int(decoded.width))x\(Int(decoded.height))→\(Int(drawable.width))x\(Int(drawable.height))|hdr\(hdrActive ? 1 : 0)"
         if sig != lastSizeSig {
             lastSizeSig = sig
+            // Explicit verdict: is the shader presenting 1:1 (decoded == drawable) or resampling? The
+            // scale ratio makes a residual match-window mismatch obvious. If this says 1:1 but the
+            // picture is still soft, the resample is downstream of us (macOS compositor — a scaled
+            // display mode, or a fractional-pixel window position), not the shader.
+            let sx = decoded.width > 0 ? drawable.width / decoded.width : 0
+            let sy = decoded.height > 0 ? drawable.height / decoded.height : 0
+            let verdict = decoded == drawable
+                ? "1:1 (no resample)"
+                : String(format: "RESAMPLE scale=%.4fx%.4f", sx, sy)
             let msg =
-                "stage2: decoded \(Int(decoded.width))x\(Int(decoded.height)) → drawable \(Int(drawable.width))x\(Int(drawable.height)) hdr=\(hdrActive)"
+                "stage2: decoded \(Int(decoded.width))x\(Int(decoded.height)) → drawable \(Int(drawable.width))x\(Int(drawable.height)) [\(verdict)] hdr=\(hdrActive)"
             presenterLog.info("\(msg, privacy: .public)")
         }
     }

clients/apple/Sources/PunktfunkKit/Video/SessionPresenter.swift

@@ -205,17 +205,33 @@ final class SessionPresenter {
             return nil
         }()
         let fit: CGRect = aspect.map { AVMakeRect(aspectRatio: $0, insideRect: bounds) } ?? bounds
+        // Snap the sublayer frame to the BACKING PIXEL GRID. AVMakeRect centers the aspect-fit rect,
+        // so its origin/size are usually fractional points; a metal sublayer whose frame doesn't land
+        // on whole device pixels is RESAMPLED by the macOS/UIKit compositor during composite — a
+        // uniform "everything looks soft" blur — even when the drawable itself is pixel-exact 1:1
+        // (verified via the stage2 "[1:1 (no resample)]" log while the picture was still soft). Round
+        // origin AND size to device pixels so the composite is a true 1:1 blit. Idempotent when the
+        // frame is already aligned (e.g. fullscreen fit == integer bounds), so it's a no-op there.
+        let scale = contentsScale > 0 ? contentsScale : 1
+        let snapped = CGRect(
+            x: (fit.origin.x * scale).rounded() / scale,
+            y: (fit.origin.y * scale).rounded() / scale,
+            width: (fit.width * scale).rounded() / scale,
+            height: (fit.height * scale).rounded() / scale)
         // No implicit resize animation; contentsScale tracks the view's backing/display scale.
         CATransaction.begin()
         CATransaction.setDisableActions(true)
         metalLayer.contentsScale = contentsScale
-        metalLayer.frame = fit
+        metalLayer.frame = snapped
         CATransaction.commit()
         // Hand the resulting pixel size to the render thread (it must not read layer geometry
-        // cross-thread) — this is what the presenter sizes its drawable to.
+        // cross-thread) — this is what the presenter sizes its drawable to. Uses the SNAPPED size so
+        // the drawable's texel count equals the on-screen device-pixel count exactly (1 texel ↔ 1
+        // device pixel); with the frame snapped, this equals the pre-snap rounded value, so the
+        // decoded↔drawable 1:1 the log confirmed is preserved.
         stage2?.setDrawableTarget(CGSize(
-            width: (fit.width * contentsScale).rounded(),
-            height: (fit.height * contentsScale).rounded()))
+            width: (snapped.width * scale).rounded(),
+            height: (snapped.height * scale).rounded()))
         #if os(tvOS)
         // Push the display's live EDR headroom alongside: > 1 means the TV is composited in an
         // HDR mode (the session's AVDisplayManager request landed — see StreamViewIOS), and HDR