RENDER-DRIVER CALLER MAP, and why no element type is ever written.

My earlier ROM scans were wrong: CALLA stores the BIT address, not the file
offset (the decode of the slot-21 call site at file 0x63098 is
  5f 0d 90 8a 2d 3f   ->  CALLA 0x3F2D8A90  =  file 0x1B152
which is the form Ghidra mis-prints as "0x07e5b152").  Scanning for bit
addresses instead of file numbers yields the real caller list.

EVERY reference into the render-driver region targets the ENTRY 0x1B152:

  caller file 06309A -> 0x1B152   (slot 21 = EXPORT)
  caller file 065186 -> 0x1B152   (slot 33 body)
  caller file 065550 -> 0x1B152   (slot 40 body)
  caller file 065BFA -> 0x1B152   (slot 43)
  caller file 065F54 -> 0x1B152   (slot 44)

No reference anywhere targets the per-element blocks (0x1B29A / 0x1B342 /
0x1B46A-entry / 0x1B4F4-entry / 0x1B582-entry).  Those are therefore reachable
ONLY by control flow inside 0x1B152 itself.

Element-type stores, one per type, all the same instruction shape:
  0x1B2B8/0x1B2BA  MOVK 0x1,A11 ; MOVB *A11,*A5(0x58)   type 1
  0x1B360/0x1B362  MOVK 0x2,A11 ; MOVB *A11,*A5(0x58)   type 2
  0x1B46A/0x1B46C  MOVK 0x3,A11 ; MOVB *A11,*A5(0x58)   type 3
  0x1B4F4/0x1B4F6  MOVK 0x4,A11 ; MOVB *A11,*A5(0x58)   type 4
  0x1B582/0x1B584  MOVK 0x5,A11 ; MOVB *A11,*A5(0x58)   type 5

CONSEQUENCE (structural): the CLIENT does not select an element type by calling
these blocks -- there is no such call.  The type is chosen inside 0x1B152 from
whatever pipeline/band data it is given.  Since our band element record is
uninitialised (word1 = 0x00000055 heap fill) and no TYPE store ever fires, the
server finds no elements to add: the pipeline is empty because the band
descriptor is empty, not because an element request is missing.

Five requests can drive the renderer: slots 21, 33, 40, 43, 44.  Slots 33/40/44
are the bind-family handlers that call both the resource lookup (0x7EA6FA0) and
the bind helper (0x7EA72A4); their operand requirements are what I have not been
able to satisfy (they reject with BadValue/BadIDChoice before reaching the
renderer).

NET: the missing capability is still the 2.3 encoding of the pipeline/band
descriptor that these five requests carry.  The call-graph is now fully mapped,
which is real structure, but it does not by itself yield a working request.
