perf(video): 消除中间帧复制并优化 MJPEG 有界预取

通过 Bytes 所有权和只读 AVBufferRef 引用池化 NV12 输入，消除中间 AVFrame 的整帧复制，并保留 V4L2 编码包脱离硬件缓冲的处理。

允许输出健康时有界预取 MJPEG，在输出背压时优先解码前丢帧；增加等帧、输出等待、丢帧原因及编码分段耗时统计，提供预取和所有权输入的对照开关。

补充缓冲寿命、只读保护、软件编码回退和队列策略测试及接口说明，保留现有 PTS、解码 worker、音频和加密行为。

libs/hwcodec/src/ffmpeg_ram/encode.rs

@@ -1,5 +1,7 @@
 #[cfg(feature = "bytes")]
-use crate::ffmpeg_ram::{ffmpeg_ram_encode_packet, ffmpeg_ram_free_packet};
+use crate::ffmpeg_ram::{
+    ffmpeg_ram_encode_owned_packet, ffmpeg_ram_encode_packet, ffmpeg_ram_free_packet,
+};
 use crate::{
     common::DataFormat::{self, *},
     ffmpeg::{init_av_log, AVPixelFormat},
@@ -498,6 +500,23 @@ pub struct Encoder {
     pub linesize: Vec<i32>,
     pub offset: Vec<i32>,
     pub length: i32,
+    owned_nv12_input: bool,
+    measure: bool,
+    #[cfg(feature = "bytes")]
+    output_copy_us: u64,
+}
+
+#[derive(Clone, Copy, Debug, Default)]
+pub struct EncodeInputStats {
+    pub input_frames: u64,
+    pub borrowed_frames: u64,
+    pub copied_frames: u64,
+    pub copied_bytes: u64,
+    pub prepare_us: u64,
+    pub send_us: u64,
+    pub receive_us: u64,
+    pub packet_us: u64,
+    pub output_copy_us: u64,
 }
 
 impl Encoder {
@@ -545,6 +564,9 @@ impl Encoder {
                 return Err(());
             }
 
+            let owned_nv12_input = ctx.name.ends_with("_v4l2m2m")
+                && ctx.pixfmt == AVPixelFormat::AV_PIX_FMT_NV12 as i32
+                && std::env::var("ONE_KVM_V4L2_OWNED_INPUT").as_deref() != Ok("0");
             Ok(Encoder {
                 codec,
                 frames: Box::into_raw(Box::new(Vec::<EncodeFrame>::new())),
@@ -552,6 +574,10 @@ impl Encoder {
                 linesize,
                 offset,
                 length: length[0],
+                owned_nv12_input,
+                measure: std::env::var("ONE_KVM_VIDEO_STATS").as_deref() == Ok("1"),
+                #[cfg(feature = "bytes")]
+                output_copy_us: 0,
             })
         }
     }
@@ -581,33 +607,107 @@ impl Encoder {
 
     #[cfg(feature = "bytes")]
     pub fn encode_bytes(&mut self, data: &[u8], ms: i64) -> Result<Vec<EncodeBytesFrame>, i32> {
+        let length = c_int::try_from(data.len()).map_err(|_| -22)?;
         unsafe {
             let mut frames = Vec::<EncodeBytesFrame>::new();
             let result = ffmpeg_ram_encode_packet(
                 self.codec,
                 data.as_ptr(),
-                data.len() as _,
+                length,
                 &mut frames as *mut _ as *const c_void,
                 ms,
                 Some(Encoder::packet_callback),
             );
-            if result == -11 || result == 0 {
-                if self.ctx.name.contains("v4l2m2m") {
-                    return Ok(frames
-                        .into_iter()
-                        .map(|frame| EncodeBytesFrame {
-                            data: Bytes::copy_from_slice(frame.data.as_ref()),
-                            pts: frame.pts,
-                            key: frame.key,
-                        })
-                        .collect());
-                }
-                return Ok(frames);
-            }
-            Err(result)
+            self.finish_bytes_encode(frames, result)
         }
     }
 
+    pub fn supports_owned_nv12_input(&self) -> bool {
+        self.owned_nv12_input
+    }
+
+    pub fn input_stats(&self) -> Option<EncodeInputStats> {
+        if !self.measure {
+            return None;
+        }
+        let mut stats = unsafe { std::mem::zeroed() };
+        unsafe { crate::ffmpeg_ram::ffmpeg_ram_encoder_stats(self.codec, &mut stats) };
+        Some(EncodeInputStats {
+            input_frames: stats.input_frames,
+            borrowed_frames: stats.borrowed_frames,
+            copied_frames: stats.copied_frames,
+            copied_bytes: stats.copied_bytes,
+            prepare_us: stats.prepare_us,
+            send_us: stats.send_us,
+            receive_us: stats.receive_us,
+            packet_us: stats.packet_us,
+            #[cfg(feature = "bytes")]
+            output_copy_us: self.output_copy_us,
+            #[cfg(not(feature = "bytes"))]
+            output_copy_us: 0,
+        })
+    }
+
+    #[cfg(feature = "bytes")]
+    pub fn encode_owned_bytes(
+        &mut self,
+        data: Bytes,
+        ms: i64,
+    ) -> Result<Vec<EncodeBytesFrame>, i32> {
+        if !self.owned_nv12_input {
+            return self.encode_bytes(data.as_ref(), ms);
+        }
+        let length = c_int::try_from(data.len()).map_err(|_| -22)?;
+        let owner = Box::new(data);
+        let pointer = owner.as_ptr();
+        let mut frames = Vec::<EncodeBytesFrame>::new();
+        let result = unsafe {
+            ffmpeg_ram_encode_owned_packet(
+                self.codec,
+                pointer,
+                length,
+                Box::into_raw(owner) as *mut c_void,
+                Some(Self::release_input),
+                &mut frames as *mut _ as *const c_void,
+                ms,
+                Some(Self::packet_callback),
+            )
+        };
+        self.finish_bytes_encode(frames, result)
+    }
+
+    #[cfg(feature = "bytes")]
+    fn finish_bytes_encode(
+        &mut self,
+        frames: Vec<EncodeBytesFrame>,
+        result: i32,
+    ) -> Result<Vec<EncodeBytesFrame>, i32> {
+        if result != -11 && result != 0 {
+            return Err(result);
+        }
+        if !self.ctx.name.contains("v4l2m2m") {
+            return Ok(frames);
+        }
+        let start = self.measure.then(std::time::Instant::now);
+        let detached = frames
+            .into_iter()
+            .map(|frame| EncodeBytesFrame {
+                data: Bytes::copy_from_slice(frame.data.as_ref()),
+                pts: frame.pts,
+                key: frame.key,
+            })
+            .collect();
+        if let Some(start) = start {
+            self.output_copy_us += start.elapsed().as_micros() as u64;
+        }
+        Ok(detached)
+    }
+
+    #[cfg(feature = "bytes")]
+    extern "C" fn release_input(owner: *mut c_void, _data: *mut u8) {
+        unsafe { drop(Box::from_raw(owner as *mut Bytes)) };
+    }
+
     extern "C" fn callback(data: *const u8, size: c_int, pts: i64, key: i32, obj: *const c_void) {
         unsafe {
             let frames = &mut *(obj as *mut Vec<EncodeFrame>);