refactor(avhw): split encoder module

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

src/avhw/types.rs

@@ -0,0 +1,88 @@
+/// Commands sent from the WebRTC thread to the SW encoder when the
+/// bandwidth estimate changes significantly.
+pub enum BitrateCommand {
+    UpdateBitrate {
+        target_bps: u64,
+    },
+    UpdateResolution {
+        width: u32,
+        height: u32,
+    },
+    /// Force the next encoded frame to be an IDR. Sent by the WebRTC thread
+    /// in response to str0m `Event::KeyframeRequest` or a resolution change.
+    ForceKeyframe,
+}
+
+#[derive(Clone, Copy, Debug)]
+pub struct ResolutionChange {
+    pub width: u32,
+    pub height: u32,
+}
+
+/// Per-frame timing snapshot for the software encoder, consumed by the stats
+/// thread. `sws_us` measures NV12->YUV420P conversion, `encode_us` measures
+/// `avcodec_send_frame` + drain, and `output_bytes` counts encoded bytes
+/// produced by libavcodec (even if downstream delivery later drops them).
+#[derive(Default, Clone, Copy, Debug)]
+pub struct SwEncodeTiming {
+    pub sws_us: u64,
+    pub encode_us: u64,
+    pub output_bytes: usize,
+}
+
+/// Outcome of a single `encode_cpu_frame` call. Used by the encode thread
+/// to decide whether to report timing stats (only real encodes tick encoded_fps).
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum EncodeOutcome {
+    /// Frame was actually encoded and produced output bytes.
+    Encoded,
+    /// Frame was dropped because WebRTC is paused (no client connected).
+    SkippedPaused,
+    /// Frame was dropped because the encoder is in disconnected state.
+    SkippedDisconnected,
+    /// Frame was dropped because its Y-plane hash matched the previous frame.
+    SkippedDuplicate,
+}
+
+/// Per-stage timing breakdown for one encode cycle on the hardware path.
+/// Returned by `EncState::encode_frame` so callers can fold the numbers
+/// into `crate::stats::FrameTimings`. `transfer_us` is always 0 on the HW
+/// path because the frame stays on the GPU; the SW path's struct (if added
+/// later) would carry a real readback measurement.
+#[derive(Debug, Default, Clone, Copy)]
+pub struct EncodeStages {
+    pub scale_us: u64,
+    pub transfer_us: u64,
+    pub encode_us: u64,
+}
+
+/// Encoded H.264 frame with timing metadata for WebRTC output.
+///
+/// MP4 file output (FrameOutput::Muxer) does NOT use this - it writes via
+/// avformat which preserves PTS internally. WebRTC output (FrameOutput::Channel)
+/// requires explicit PTS propagation so RTP timestamps reflect real capture time.
+/// Without this, WebRTC clients' jitter buffers grow to seconds under
+/// damage-driven variable frame rate. See issue #24.
+#[derive(Debug)]
+pub struct EncodedH264Frame {
+    /// H.264 NAL byte stream (Annex B or AVCC depending on encoder configuration)
+    pub data: Vec<u8>,
+    /// PTS in encoder time_base units (1/fps seconds), normalized so first frame = 0.
+    /// Derived from real capture time, NOT frame counter.
+    pub pts_ticks: i64,
+    /// Wall-clock capture time, propagated from CpuNv12Frame for frame_age stat.
+    pub capture_time: std::time::Instant,
+}
+
+/// Owned CPU NV12 frame data for cross-thread transfer.
+/// Produced by main thread (VAAPI import + GPU scale + transfer), consumed by encode thread.
+pub struct CpuNv12Frame {
+    pub y_data: Vec<u8>,
+    pub uv_data: Vec<u8>,
+    pub y_stride: usize,
+    pub uv_stride: usize,
+    pub pts: i64,
+    /// Wall-clock time when this frame was captured (PipeWire delivery).
+    /// Used for frame_age stat: time from capture to WebRTC send.
+    pub capture_time: std::time::Instant,
+}