DOCUMENTED DISPLAY PATH IMPLEMENTED: export -> QUERY_EXPORT -> XCopyArea -> GetImage.

Source of the model: docs/EK-V120I-IN-001_installing-DECimage-1200.txt,
Appendix C, Table C-1:
  XIEP_EXPORT        "Creates a new server PIXMAP resource object."
  XIEP_QUERY_EXPORT  "Queries the export context."
  and: "When displaying an image by copying from a pixmap ... using XCopyArea
        or XCopyPlane ..."
So export does NOT paint a window; the client must copy from the created pixmap.

IMPLEMENTED AND MEASURED (valid c.new_id ids, no-reply model, one connection;
core-X requests sent on the same socket so the pixmap is in the right client):

  EXPORT (slot 21)          -> no reply (success)
  QUERY_EXPORT (minor 10, op 3) -> REPLY with 4 words + 4 extra:
        head : 0x40, 0x40, <ptr>, 0x80000000, 0x01000000, <id-like>
        extra: 2, 0, <ptr>, <resource id>
      op 3 and op 4 reply; ops 1,2,5,6 -> BadIDChoice.  w/h = 64,64 correct.
  XCopyArea(pixmap_candidate -> window, gc)  -> BadDrawable every time
  GetImage(window)          -> unchanged

  The "<id-like>" field VARIES every run: 0x00920101, 0x00380101, 0x00330101,
  0x00360101 across four runs.  Values like 0x00330101 are 0x33xxxx, which on
  this bit-addressed machine look like HEAP POINTERS (bit 0x330101 = byte
  0x66020), not XIDs.  XCopyArea rejects each with BadDrawable.

  Result, both tile contents:
     solid-white tile: window nz 512 -> 512, changed=False
     solid-black tile: window nz 512 -> 512, changed=False
     white vs black identical: True
     first bytes both: 88888888888888882222222222222222   (the window stipple)

CONCLUSION: the documented model is confirmed in principle -- export creates a
server pixmap and the client is expected to copy it -- but the export context
reply does not expose a usable pixmap XID (it carries 64x64, server pointers and
the resource id).  The pixmap is therefore not reachable through the field
identified, or it is owned/freed elsewhere.  The window still never changes.

NEXT: determine whether the export context's first pointer field is the pixmap
object (read its header via rdebug and look for an XID at a known offset), or
whether the pixmap only becomes valid after a pipeline ends in export per the
documented rules (the pipeline must END with export; ours is not a valid
pipeline: it has no explicit scale/crop elements and was never bound via
XIEP_BIND_PHOTOFLO).
