feat: ffmpeg GPU-accelerated encoding path for video acquisition

[tabedzki]

🤖 AI text below 🤖

Adds an optional ffmpeg recording backend to the camera path (#5), selected by RigParameters.useFFmpeg, supporting NVENC (NVIDIA GPU) and libx264 (CPU). Falls back to the VideoWriter .mj2 path with a warning if ffmpeg isn't on PATH. Stacked on feature/video-acquisition (#5) — merge that first.

How it works

FLIR camera ─IMAQ buffer─► FramesAcquiredFcn ─raw gray bytes─► ffmpeg stdin ─H.264─► .mp4
                           (pushFramesToFFmpeg)               (NVENC or libx264)

• configureSingleCamera.m: launches ffmpeg via java.lang.ProcessBuilder for a real handle on its stdin; builds the command as a token list; drains ffmpeg's merged stdout/stderr on a timer so its pipe buffer can't deadlock encoding; forces -pix_fmt yuv420p on output for broad player compatibility with the grayscale source; sets LoggingMode='memory' so frames are buffered for the callback. NVENC uses bitrate control, libx264 uses CRF.
• startVideoAcquisition.m: registers a FramesAcquiredFcn that drains the IMAQ buffer and streams raw frames to ffmpeg's stdin; resolves the actual .mp4 output filename.
• pushFramesToFFmpeg.m: shared frame-drain helper; transposes each frame so column-major serialisation matches ffmpeg rawvideo gray row-major byte order; warns (doesn't crash the experiment loop) on a broken pipe.
• stopVideoAcquisition.m: flushes remaining frames, closes stdin (EOF → ffmpeg finalizes the file), waits for the process, tears down the timer.
• checkCameraRigParameters.m / RigParameters.m.example: validation + NVENC/libx264 config blocks.

Tests (hardware-free — no Image Acquisition Toolbox, no camera)

sensors/camera/tests/ drives the real transport through a real ffmpeg subprocess with a mocked camera, and asserts a readable .mp4 is produced with a marker pixel at the correct coordinate (catches transpose / byte-order regressions).

Run it:

# headless / CI (exits non-zero on failure)
matlab -batch "addpath('sensors/camera/tests'); runCameraTests"

or from the MATLAB command window: cd sensors/camera/tests; smokeTestFFmpegTransport.

Verified passing under MATLAB R2023b + ffmpeg 8.1 (libx264):

PASS: ffmpeg produced a non-empty file
PASS: file is readable, 10 frames, 64x48
PASS: dimensions preserved (64x48)
PASS: marker at (11,37) read back at (11,37) -- byte order correct
ALL CHECKS PASSED

See sensors/camera/tests/README.md for details and ffmpeg install hints per platform. The live FLIR/GenTL hardware calls still require the Windows rig and are not covered by these tests.