DECISIVE NEGATIVE: NOTHING FILL-DEPENDENT REACHES THE STAGING BUFFER OR VRAM.

Watched writes into the staging buffer 0x004FE780..0x004FE97F for a WHITE export
and for a BLACK export, in the same session:

  fill=ff : 802 writes, values: 0000 x310, FFFF x250, 0001 x37, 5555 x32
  fill=00 : 802 writes, values: 0000 x310, FFFF x250, 0001 x37, 5555 x32
                                                   ^^^^ IDENTICAL

The staging traffic is byte-for-byte the same for white and black.  So it does
not carry the client's tile -- the 0xFFFF words are engine/Terminal-Manager
constants, not our fill.  Combined with the previous turn's VRAM finding (also
identical for both fills), the conclusion is:

  THE ONLY FILL-DEPENDENT DATA IN THE WHOLE SYSTEM IS THE CLIENT-SIDE IMPORT
  (element+0x28 = 0xFF for white, 0x00 for black).  Everything downstream of the
  accelerator FIFO -- the engine output port, the staging buffer and VRAM -- is
  fill-independent, i.e. our model of the engine's output path is not delivering
  the pixels, and nothing else downstream copies them either.

CUMULATIVE, HONEST STATE
  * client/protocol/library: complete (DEC 2.3 built from DEC's own sources);
  * import: verified and fill-dependent;
  * export: completes without error, opt-in model only;
  * accelerator FIFO: receives 0xFFFF/0x0000 (fill-dependent) -- the last point
    where the data is proven to be ours;
  * engine output 0x5600200 -> staging 0x004FE790 -> VRAM: all fill-INDEPENDENT,
    so the tile is lost somewhere in that modelled path;
  * NO client-chosen XIE image has ever been displayed;
  * default configuration unchanged; xshow (core X) remains the only viewer that
    has shown an image.
