The embedded FLV videos were encoded like this:
  $ ffmpeg -i gray_on_black.png -q 0 gray_on_black_0.flv
And similarly for `q` values of 11, 22, and 33; and for the other two images.

Since the test images consist only of constant colored blocks (8x8 for luma and
16x16 for chroma, due to subsampling), the `q` option doesn't actually change the
encoded image at all. The decoder, however, doesn't know this, and will more eagerly
smooth out the block edges of the same picture if it says it was quantized more.

What this test DOES check:
 - Which deblocking algorithm is in use and how it affects the luma and chroma planes.
 - How its strength is affected by the encoded quality/quantizer setting.

What it DOES NOT check:
 - How the deblocking strength of non-first (for example, inter-) frames is selected.
 - How/if the deblocking of one frame affects any neighboring frames.
 - Exactly when deblocking is used or not, based on: playback quality, frame count
   of the video, stream or frame tag flags, etc.
