feat: 新增 Amlogic AMLENC 硬件编码支持

- 新增动态加载及 ABI 校验的 AMLENC H.264/H.265 编码器
- 注册 Amlogic 后端并扩展编码器自检
- 并行执行 MJPEG 解码与 AMLENC 编码
- 复用 NV12 缓冲区并移除冗余帧初始化
- 优化低延迟帧分发及 RTCP 关键帧恢复
- 动态调整 AMLENC 码率尚未完成，效果不佳

res/vcpkg/libyuv/src/lib.rs

@@ -1119,25 +1119,73 @@ pub fn mjpg_size(src: &[u8]) -> Result<(i32, i32)> {
 
 /// Decode MJPEG directly to NV12.
 pub fn mjpg_to_nv12(src: &[u8], dst: &mut [u8], width: i32, height: i32) -> Result<()> {
-    if width % 2 != 0 || height % 2 != 0 {
-        return Err(YuvError::InvalidDimensions);
-    }
-
-    let w = width as usize;
-    let h = height as usize;
-    if dst.len() < nv12_size(w, h) {
+    let (y_size, output_size) = mjpg_nv12_plane_sizes(width, height)?;
+    if dst.len() < output_size {
         return Err(YuvError::BufferTooSmall);
     }
 
-    let y_size = w * h;
     let (dst_y, dst_uv) = dst.split_at_mut(y_size);
+    // SAFETY: the length check above guarantees writable storage for both planes.
+    unsafe { mjpg_to_nv12_raw(src, dst_y.as_mut_ptr(), dst_uv.as_mut_ptr(), width, height) }
+}
 
+/// Decode MJPEG directly into a reusable `Vec` without zero-filling the output first.
+///
+/// `Vec::resize` must initialize every byte before libyuv immediately overwrites the
+/// complete NV12 frame. This variant lets libyuv initialize spare capacity directly
+/// and publishes the new length only after a successful conversion.
+pub fn mjpg_to_nv12_vec(src: &[u8], dst: &mut Vec<u8>, width: i32, height: i32) -> Result<()> {
+    let (y_size, output_size) = mjpg_nv12_plane_sizes(width, height)?;
+
+    dst.clear();
+    dst.reserve(output_size);
+
+    // SAFETY: reserve above guarantees writable capacity for the Y and UV planes.
+    // MJPGToNV12 writes the complete output on success; set_len is deliberately
+    // delayed until then so callers can never observe partially initialized bytes.
+    let result = unsafe {
+        let dst_y = dst.as_mut_ptr();
+        mjpg_to_nv12_raw(src, dst_y, dst_y.add(y_size), width, height)
+    };
+    result?;
+
+    // SAFETY: a successful MJPGToNV12 call initialized exactly output_size bytes.
+    unsafe { dst.set_len(output_size) };
+    Ok(())
+}
+
+#[inline]
+fn mjpg_nv12_plane_sizes(width: i32, height: i32) -> Result<(usize, usize)> {
+    if width % 2 != 0 || height % 2 != 0 || width <= 0 || height <= 0 {
+        return Err(YuvError::InvalidDimensions);
+    }
+    let y_size = (width as usize)
+        .checked_mul(height as usize)
+        .ok_or(YuvError::InvalidDimensions)?;
+    let output_size = y_size
+        .checked_mul(3)
+        .map(|size| size / 2)
+        .ok_or(YuvError::InvalidDimensions)?;
+    Ok((y_size, output_size))
+}
+
+/// # Safety
+///
+/// `dst_y` and `dst_uv` must point to writable planes sized for `width` x `height` NV12.
+#[inline]
+unsafe fn mjpg_to_nv12_raw(
+    src: &[u8],
+    dst_y: *mut u8,
+    dst_uv: *mut u8,
+    width: i32,
+    height: i32,
+) -> Result<()> {
     call_yuv!(MJPGToNV12(
         src.as_ptr(),
         usize_to_size_t(src.len()),
-        dst_y.as_mut_ptr(),
+        dst_y,
         width,
-        dst_uv.as_mut_ptr(),
+        dst_uv,
         width,
         width,
         height,