TURN RECORD: THE TRANSPORT PATH IS NOW FULLY CLIENT-DRIVEN; EXPORT RUNS THE
RENDERER BUT STILL FINDS NO SOURCE.

A) XieCreatePhoto requirement discovered by measurement
   The create request is XieCreatePhoto(type, width, height, cmp_map, cmp_cnt,
   cmp_lvl[], pxl_pol, pxl_ratio, pxl_prog, scn_prog) -- man page
   xie/docs/dec/xie23/man/XieCreatePhoto.3X.  Sweeping the words that follow
   width/height (w4..w8, unique id per trial, full pipeline each time):

     baseline (w4..w8 = 0)  transport=err8  puttile=err10  (fails)
     w4=1                   err8 / err10
     w5=1                   err8 / err10
     ** w6=1                transport=ok/silent  puttile=ok/silent **
     w7=1                   err8 / err10
     w8=1                   err8 / err10
     w5=0x020202            err8 / err10
     w6=1 w5=0x020202       ok / ok

   So **w6 = 1 is required** (the component count / image-attribute word).
   Without it the photomap has no components, SetTransport fails with BadMatch
   and PutTile with BadAccess.  With it, and with SetTransport mode=3
   (component_organization=4, plane_mask=1, pixel_stride[0]=1,
   scanline_stride[0]=8), the whole client path works:

     InitSession(2.3) -> CreatePhoto(w6=1) -> SetTransport(mode=3)
        -> PutTile(64x64, plane_mask=1, final=1)   = ok/silent, NO DEBUG POKES

B) EXPORT FROM AN UNBOUND PHOTOMAP RUNS THE RENDERER BUT HAS NO SOURCE
     export(minor 21, 16 words, w=64 h=64) -> ok/silent
     trace: RP(0x1B296) fires with src_base610=00000000, and the TYPE5 element
     store executes.
   w=0,h=0 skips the renderer entirely.
   Framebuffer captures (dwt/screenshot.py) for a solid-white tile and a
   solid-black tile differ by 1600-1608 pixels -- the Terminal Manager's memory
   figures and the pointer cursor.  The 64x64 window itself is unchanged.

C) THE BOUND PATH DOES NOT WORK AS-IS
   Create a photoflo (type=1, w6=1), BindPhotomap (11) with [photomap, photoflo]
   -> ok/silent, then Export(21) -> err14 (BadIDChoice), ExecutePhotoflo(13)
   -> err1.  Per XiePutTile.3X a photomap bound to a flo can no longer receive
   tiles, so this ordering is wrong: tiles must go in BEFORE bind, which is what
   B does, and then the unbound-immediate export is the right request -- yet
   src_base stays 0.

D) QUERYEXPORT IS NOT USABLE YET
   minor 24 with 3 and 4 words (photo_id alone, photo_id+drawable) always
   returns err1 (BadRequest), so the export context cannot be inspected to see
   whether a pixmap was produced.

E) CORRECTION FROM THE PEER SESSION, checked against the header
   XieFloatRec is NOT an IEEE float (XieFloat.h: {u8 e_sign; u8 m_sign;
   u16 exponent; u32 mantissa}).  Zero-filling match_limit/gray_limit encodes
   mantissa=0 => 0.0, which is a valid value ("exact matches only"), so that was
   not the failure.  Encoding 1.0 would be mantissa 0x80000000, exponent 1.

STATUS.  Still NO client-chosen XIE image.  src_base (rp+0x610) remains
00000000 in every run; the window is unchanged; solid-white and solid-black
captures are identical apart from the Terminal Manager's own stats; xshow
(core X) remains the only viewer that has displayed an image.

WHAT IS NOW TRUE THAT WAS NOT BEFORE: the import half of the path is entirely
client-driven and reproduces (no pokes), and the create request has a known,
required attribute word (w6=1).  The remaining gap is only the export half: the
renderer runs but reports no source, so either the photomap is still not
"populated" in the sense XieExport needs, or the export needs a length/element
form I have not matched.
