ATTRIBUTION: THE VRAM WRITES ARE NOT THE IMAGE.

Measured with a write watch over VRAM 0x04000000..0x0403FFFF during a completing
export under VT1200_ACCEL=1:

  idle (6 s, no client)                 :      0 writes
  total after one full export cycle     : 43,882 writes   (peak 0x0403FCF0)
  writes attributable to the export call:    241

So the ~43.9k writes happen when the client's window is created/mapped and the
Terminal Manager repaints -- not from the pixel engine.  The export itself
contributes only 241 words, which is nowhere near the 64x64 tile (4096 pixels /
512 bytes / 256 words at minimum), and the framebuffer is unchanged:

  white vs black captures differ by AE 1841-2600 px = TM figures + cursor only.

CORRECTING AN EARLIER READING: "43,882 VRAM writes with 0xFFFF during the export"
was wrong -- those writes are the window/TM redraw.  The export's own VRAM
footprint is 241 words and does not draw the tile.

TRIED THIS TURN
  * watch VRAM during export, with the renderer at pitch 2048 (default VR319) AND
    at pitch 1024 (VT1200_MONITOR=VR150) in case the write address mapped
    off-screen; neither shows the block.
  * the accelerator FIFO does receive the source pixels (verified previously:
    2200/4400 writes containing 0xFFFF / 0x0000), so the data path is real; the
    engine simply never performs the store.

STATE
  * export completes without error (opt-in model);
  * source pixels reach the accelerator FIFO;
  * the engine performs no VRAM store; destination/placement is still undecoded;
  * NO client-chosen XIE image displayed;
  * model opt-in (VT1200_ACCEL=1); default configuration unchanged;
  * xshow (core X) remains the only viewer that has shown an image.
