Output: pick the codec from the extension; add --keep-audio

[joeljose]

Closes #43.

Changes

• The extension picks the codec:
  • .mkv writes FFV1, lossless; the frames read back match the 8-bit output exactly (the acceptance test);
  • .avi writes MJPG (unchanged, still the default output);
  • .mp4 writes MPEG-4 Part 2 (mp4v). H.264 isn't available in OpenCV's pip builds.
  • Any other extension is an error before processing starts. Before, any container got MJPG, with a warning.
• --keep-audio copies the input's audio track into the output with ffmpeg -map 0:v -map 1:a? -c copy. Without ffmpeg it warns at startup and writes video only. If the container can't hold the audio codec, it warns and keeps the video-only file.
• The README flag table recommends .mkv for analysis: JPEG compression noise is as large as the sub-1% colour changes EVM reveals.

Not done

• PNG sequences and --save-npy: suggested, but not in the acceptance criteria. .mkv already gives lossless 8-bit output, and library users get the float result from eulerian_magnification().
• ffmpeg in the Docker images: it would add weight for an optional flag. The images warn and keep the video.

Verification

• ./test.sh: all pass. test_keep_audio_end_to_end is skipped there because the image has no ffmpeg.
• Locally with ffmpeg 6.1: 104 passed, 1 skipped, including an end-to-end --keep-audio run (ffprobe shows a video and an audio stream).
• Manual check with face.mp4 plus an AAC track: .mkv gives ffv1 + aac, .mp4 gives mpeg4 + aac, .avi gives mjpeg + aac. -o out.webm exits 1 with the list of supported extensions.