MILESTONE: THE XIE EXPORT REQUEST NOW COMPLETES WITHOUT ERROR.

With VT1200_ACCEL=1 and the accelerator status model reporting
  bit 1  initialised        (polled at file 0x122E54)
  bit 2  transfer complete  (polled at 0x125448)
  bit 9  done               (polled at 0x1253FC and on other control registers)
  bit 0  data ready, as a LATCH -- raised by a FIFO command write (0x5600080),
         cleared by the first FIFO data write (0x5600050)

the client reports, for BOTH fills:

  display 192.168.99.2:0 opened (root depth 1)
  image: org=3 w=64 h=64 levels=2,0,0 arSize=4096
  photomap created
  put tile sent
  export sent (with colormap)          <-- NO X ERROR

Previously every export returned BadValue(2) or BadMatch(8).  The request is now
accepted and the whole DDX render path runs to completion.

WHAT IS STILL MISSING
  The framebuffer does not change: white vs black captures differ by ~2486 px,
  which is the Terminal Manager's own figures and the cursor, not the 64x64
  window.  The accelerator model reports status honestly but performs no pixel
  transfer -- it has no idea of destination, geometry or data format, and the
  154-write command stream to 0x5600080 has not been decoded.

  So the protocol layer is now complete end to end; what remains is the pixel
  engine itself.

STATE
  * default configuration unchanged and verified (model is opt-in).
  * NO client-chosen XIE image displayed yet.
  * xshow (core X) remains the only viewer that has shown an image.

NEXT: decode the 0x5600080 command stream (154 writes captured earlier) to learn
destination address/geometry, then implement the FIFO -> framebuffer copy in
accel_w so the transfer actually paints.
