fresnel-fourier iter5b Phase 6 commit B: state-tracking — request.h field + config.c wire-up

request.h: add last_output_pixelformat to struct request_data, alongside
the existing last_output_{width,height} V4L2 device state cache. Gates
re-S_FMT on codec change in addition to resolution change.

config.c::RequestCreateConfig: wire up object_config->pixelformat
(previously dead field at config.h:46) by calling pixelformat_for_profile
on the active profile. The pixelformat field becomes the source of truth
that surface.c reads in commit C.

Signed-off-by: claude-noether <claude-noether@reauktion.de>

src/request.h

@@ -96,9 +96,23 @@ struct request_data {
 	 * probes with small surfaces then re-allocates at real
 	 * resolution; we re-set the OUTPUT format whenever this pair
 	 * changes).
+	 *
+	 * iter5b: extended with last_output_pixelformat so the gate
+	 * also fires on codec change (consumer decodes H.264 first,
+	 * HEVC second on the same driver_data). Pre-iter5b the OUTPUT
+	 * pixel format was hardcoded V4L2_PIX_FMT_H264_SLICE in
+	 * CreateSurfaces2, so codec change was silent and HEVC / VP9 /
+	 * VP8 saw the kernel hantro driver substitute MPEG2_DECODER
+	 * codec_mode (when bound to hantro) or rkvdec silently drop
+	 * the decode (when bound to rkvdec). Same bug class as iter4's
+	 * unconditional h264_start_code; both fixes thread the active
+	 * profile into codec-specific kernel state.
+	 *
+	 * 0 = uninitialized; non-zero = FOURCC of current OUTPUT format.
 	 */
 	unsigned int last_output_width;
 	unsigned int last_output_height;
+	unsigned int last_output_pixelformat;
 };
 
 VAStatus VA_DRIVER_INIT_FUNC(VADriverContextP context);