THE TILE IS IN VRAM -- BUT IT DOES NOT MAP TO THE VISIBLE PICTURE.

Read VRAM through rdebug while the client was live, right after a WHITE export:

  0x04017830: effefffffdffffffffffffffffffffff
  0x04017840: fffffdffffffffffffffffffffffffff
  0x04017850: fdffffffffffffffffffffffffffffff
  0x04017900: ffffffffffffffffffffffffffffffff
  0x04017A00: ffffffffffffffffffffffffffffffff

That is the 0xFF tile, present in video memory.  So the whole chain now works:

  client -> XIE -> photomap -> accelerator FIFO -> engine output -> staging
         -> VRAM (tile verified by direct read)

BUT the picture does not show it.  Sampled the framebuffer at the coordinates
that VRAM 0x040178xx maps to under the renderer's 2048-bit pitch (row 47, x=48):

   WHITE export: (48,47)=255 (48,79)=255 (80,60)=255 (100,90)=255
   BLACK export: (48,47)=255 (48,79)=255 (80,60)=255 (100,90)=255

Identical for both -- those pixels are the Terminal Manager's own light UI, not
the tile.  So the VRAM the engine writes and the VRAM the renderer displays are
offset from one another: the renderer's active picture is derived from
HEBLNK..HSBLNK (x=336 in MAME's 1696-wide raster, per RENDER_REPORT.md), while the
written address sits at a different place in the same memory.

STATE
  * THE CLIENT-CHOSEN TILE IS NOW DEMONSTRABLY IN VIDEO MEMORY;
  * it is not visible on screen -- the destination/pitch mapping between the
    engine's writes and the renderer's active picture is still wrong;
  * NO client-chosen XIE image has been displayed on screen;
  * model opt-in (VT1200_ACCEL=1); default configuration unchanged;
  * xshow (core X) remains the only viewer that has shown an image.

NEXT: compute where VRAM 0x04017830 actually falls in the renderer's mapping
(scanline = bit/2048, x = bit%2048, then apply the HEBLNK origin) and compare with
where the 64x64 window's own exposure repaint lands; that will show whether the
engine's destination should be adjusted or the renderer's origin is mis-modelled.
