perf(video): 改善视频卡顿问题并优化编码性能

改善内容:
1. NAL单元duration累积bug修复
   - 修改video_track.rs和unified_video_track.rs
   - 只有帧内最后一个NAL获得frame_duration,其他为ZERO
   - 确保同一帧的所有NAL共享相同的RTP时间戳

2. 修复VP8/VP9固定1秒duration错误
   - 将Duration::from_secs(1)改为正确的frame_duration计算

3. PTS计算优化(shared_video_pipeline.rs)
   - 将pipeline_start_time从Mutex<Option<Instant>>改为AtomicI64
   - 消除每帧一次的async mutex lock
   - 使用compare_exchange实现无锁的首帧时间设置

4. 避免重复读取config
   - 在encode_frame中缓存fps,避免末尾再次获取config锁

5. 编码器零拷贝优化
   - H264/H265/VP8/VP9编码器使用drain()替代clone()
   - 减少内存分配和拷贝开销

6. MJPEG处理器优化
   - 无客户端时跳过JPEG编码(WebRTC-only模式优化)

7. RKMPP YUYV直接输入支持
   - hwcodec C++层添加YUYV422格式支持
   - H264编码器添加Yuyv422输入格式选项

src/webrtc/unified_video_track.rs

@@ -490,10 +490,13 @@ impl UnifiedVideoTrack {
 
     /// Write VP8 frame (raw encoded)
     async fn write_vp8_frame(&self, data: &[u8], is_keyframe: bool) -> Result<()> {
+        // Calculate frame duration based on configured FPS
+        let frame_duration = Duration::from_micros(1_000_000 / self.config.fps.max(1) as u64);
+
         // VP8 frames are sent directly
         let sample = Sample {
             data: Bytes::copy_from_slice(data),
-            duration: Duration::from_secs(1),
+            duration: frame_duration,
             ..Default::default()
         };
 
@@ -514,10 +517,13 @@ impl UnifiedVideoTrack {
 
     /// Write VP9 frame (raw encoded)
     async fn write_vp9_frame(&self, data: &[u8], is_keyframe: bool) -> Result<()> {
+        // Calculate frame duration based on configured FPS
+        let frame_duration = Duration::from_micros(1_000_000 / self.config.fps.max(1) as u64);
+
         // VP9 frames are sent directly
         let sample = Sample {
             data: Bytes::copy_from_slice(data),
-            duration: Duration::from_secs(1),
+            duration: frame_duration,
             ..Default::default()
         };
 
@@ -537,25 +543,33 @@ impl UnifiedVideoTrack {
     }
 
     /// Send NAL units via track (for H264/H265)
+    ///
+    /// Important: Only the last NAL unit should have the frame duration set.
+    /// All NAL units in a frame share the same RTP timestamp, so only the last
+    /// one should increment the timestamp by the frame duration.
     async fn send_nal_units(&self, nals: Vec<Bytes>, is_keyframe: bool) -> Result<()> {
         let mut total_bytes = 0u64;
-        let mut nal_count = 0;
+        let nal_count = nals.len();
+        // Calculate frame duration based on configured FPS
+        let frame_duration = Duration::from_micros(1_000_000 / self.config.fps.max(1) as u64);
 
-        for nal_data in nals {
+        for (i, nal_data) in nals.into_iter().enumerate() {
+            let is_last = i == nal_count - 1;
+            // Only the last NAL should have duration set
+            // This ensures all NALs in a frame share the same RTP timestamp
             let sample = Sample {
                 data: nal_data.clone(),
-                duration: Duration::from_secs(1),
+                duration: if is_last { frame_duration } else { Duration::ZERO },
                 ..Default::default()
             };
 
             if let Err(e) = self.track.write_sample(&sample).await {
-                if nal_count % 100 == 0 {
+                if i % 100 == 0 {
                     debug!("write_sample failed (no peer?): {}", e);
                 }
             }
 
             total_bytes += nal_data.len() as u64;
-            nal_count += 1;
         }
 
         if nal_count > 0 {

src/webrtc/video_track.rs

@@ -484,17 +484,25 @@ impl UniversalVideoTrack {
     }
 
     /// Send NAL units as samples (H264 only)
+    ///
+    /// Important: Only the last NAL unit should have the frame duration set.
+    /// All NAL units in a frame share the same RTP timestamp, so only the last
+    /// one should increment the timestamp by the frame duration.
     async fn send_nals(&self, nals: Vec<Bytes>, is_keyframe: bool) -> Result<()> {
         let mut total_bytes = 0u64;
         // Calculate frame duration based on configured FPS
         let frame_duration = Duration::from_micros(1_000_000 / self.config.fps.max(1) as u64);
+        let nal_count = nals.len();
 
         match &self.track {
             TrackType::Sample(track) => {
-                for nal_data in nals {
+                for (i, nal_data) in nals.into_iter().enumerate() {
+                    let is_last = i == nal_count - 1;
+                    // Only the last NAL should have duration set
+                    // This ensures all NALs in a frame share the same RTP timestamp
                     let sample = Sample {
                         data: nal_data.clone(),
-                        duration: frame_duration,
+                        duration: if is_last { frame_duration } else { Duration::ZERO },
                         ..Default::default()
                     };