feat(presenter): VAAPI dmabuf → Vulkan zero-copy import + CSC pass (phase 2)

The decoder's NV12 dmabuf imports per-plane (R8 + GR88, explicit DRM
format modifier, dedicated dup'd-fd import, FOREIGN→graphics acquire)
and a fullscreen-triangle render pass converts it into the presenter's
video image with the CICP-driven coefficients ported from video_gl.rs
(same tests, plus a rows-vs-matrix agreement check). SPIR-V is committed
(shaders/build.sh regenerates) so builds and CI need no toolchain. The
import extension set is probed at device creation; unsupported boxes and
3-failure streaks demote the decoder to software via the existing
force_software contract. The session binary now honors the Settings
decoder preference instead of forcing software.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

crates/pf-presenter/shaders/build.sh

@@ -0,0 +1,10 @@
+#!/usr/bin/env bash
+# Recompile the presenter's shaders to the committed .spv blobs. The blobs are committed
+# so neither developer builds nor CI need a SPIR-V toolchain (glslc is a dev-time tool);
+# rerun this after editing a .vert/.frag and commit both.
+set -euo pipefail
+cd "$(dirname "$0")"
+for f in *.vert *.frag; do
+    glslc -O "$f" -o "$f.spv"
+    echo "compiled $f -> $f.spv"
+done

crates/pf-presenter/shaders/fullscreen.vert

@@ -0,0 +1,14 @@
+// The bufferless fullscreen triangle (same trick as the GL presenter's gl_VertexID
+// geometry). Vulkan clip space: y=-1 is the TOP, so uv (0,0) lands on the top-left —
+// which is exactly where the video's row 0 lives. No flip anywhere.
+//
+// Regenerate: shaders/build.sh (committed .spv, no build-time toolchain).
+#version 450
+
+layout(location = 0) out vec2 v_uv;
+
+void main() {
+    vec2 uv = vec2((gl_VertexIndex << 1) & 2, gl_VertexIndex & 2);
+    v_uv = uv;
+    gl_Position = vec4(uv * 2.0 - 1.0, 0.0, 1.0);
+}

crates/pf-presenter/shaders/nv12_csc.frag

@@ -0,0 +1,36 @@
+// NV12 → RGBA with the stream's CICP signaling — the Vulkan port of the GL presenter's
+// fragment shader (video_gl.rs). The YUV→RGB matrix + range expansion arrive as three
+// push-constant rows precomputed on the CPU (csc.rs `csc_rows`): rgb[i] = dot(r_i.xyz,
+// yuv) + r_i.w — one shader for BT.601/709/2020, full/limited. The chroma plane is
+// half-res; the linear sampler interpolates, same as the GL path.
+//
+// Regenerate: shaders/build.sh (committed .spv, no build-time toolchain).
+#version 450
+
+layout(location = 0) in vec2 v_uv;
+layout(location = 0) out vec4 frag;
+
+layout(set = 0, binding = 0) uniform sampler2D u_y;
+layout(set = 0, binding = 1) uniform sampler2D u_c;
+
+layout(push_constant) uniform Csc {
+    vec4 r0;
+    vec4 r1;
+    vec4 r2;
+} pc;
+
+void main() {
+    vec3 yuv = vec3(texture(u_y, v_uv).r, texture(u_c, v_uv).rg);
+    frag = vec4(
+        clamp(
+            vec3(
+                dot(pc.r0.xyz, yuv) + pc.r0.w,
+                dot(pc.r1.xyz, yuv) + pc.r1.w,
+                dot(pc.r2.xyz, yuv) + pc.r2.w
+            ),
+            0.0,
+            1.0
+        ),
+        1.0
+    );
+}