feat: known-good CPU audio + 1080p VAAPI + KMS display wall

Working snapshot before the OpenVINO MetricGAN GPU spike:
thread-capped SpeechBrain DelaySum+MetricGAN on CPU, hop 200 ms,
video hold for A/V sync, Arc VAAPI H.264, headless kmssink wall.
32 tests passing. Do not treat this as a GPU audio path.

.env.example

@@ -12,10 +12,15 @@ PARTICIPANT_PREFIX=cam
 # Streams
 # "all" = every discovered camera, "1,3,5" or "0-19" = subset
 CAMERAS=all
-VIDEO_WIDTH=1280
-VIDEO_HEIGHT=720
+VIDEO_WIDTH=640
+VIDEO_HEIGHT=360
 VIDEO_FPS=30
 VIDEO_MIN_FPS=5
+VIDEO_BITRATE=800000
+VIDEO_CODEC=h264
+VIDEO_ENCODER=vaapi
+LIBVA_DRM_DEVICE=/dev/dri/renderD129
+LIBVA_DRIVER_NAME=iHD
 
 # Audio
 AUDIO_SAMPLE_RATE=16000
@@ -44,9 +49,18 @@ PUBLISH_TIMEOUT_S=60
 LOG_LEVEL=INFO
 
 # Headless display wall (run_displays.py / livekit-displays.service)
-# 3 DRM outputs showing LiveKit participants via GStreamer kmssink
+# 3 DRM roles: 20-cam grid, designated speaker camera, screenshare STUB
 DISPLAY_COUNT=3
-DISPLAY_IDENTITIES=cam-01,cam-02,cam-03
+DISPLAY_ROLES=grid,speaker,screenshare
 # Empty = auto (connected heads first, prefer Arc HDMI-A-*)
 DISPLAY_CONNECTORS=
 DISPLAY_IDENTITY=display-wall
+# Speaker camera pane: cam-NN or "active"
+DISPLAY_SPEAKER_CAMERA=cam-01
+DISPLAY_GRID_COLS=5
+DISPLAY_GRID_ROWS=4
+DISPLAY_GRID_WIDTH=1920
+DISPLAY_GRID_HEIGHT=1080
+DISPLAY_GRID_FPS=30
+# Screen share is a stub until production LiveKit is online.
+LIVEKIT_PUBLIC_URL=