BAND / ELEMENT STRUCTURE DUMPED, BEFORE AND AFTER A BIND ATTEMPT.

Setup: valid c.new_id ids; object pointers taken from the REGIDSTORE breakpoint
(fix: the log line starts "REGIDSTORE", not "IDSTORE" -- my earlier anchored
grep silently matched nothing).

  IMG = 00403000 -> obj 003FD110  class=2  b8=1  b9=1  p44=003DC4B0
  FLO = 00404000 -> obj 00382B90  class=1  b8=0x80 b9=0x32 p44=00386510

The IMAGE resource ALREADY has an element pointer at band_array+0xd8, before any
bind:

  band[] at 003DC4B0: 00000000 00000000 00050029 40002005 48000000 48003DC5
  elem  = *(003DC4B0 + 0xd8) = 003FD7D0
  elem[]: 01050001 00000002 00100000 00000100 00004000 00000000

  (0x01050001 as bytes = 01 00 05 01; then 2; then 0x00100000; then 0x100 = 256;
   then 0x4000 = 16384.  256 and 16384 are plausibly pitch/geometry values.)

The PHOTOflo (class 1) has no element (*(p44+0xd8) reads 0x55555555, the heap
fill), and its band array is two pointers: 00382CD0 00382F50.

BIND attempts:
   slot 11 [img,flo] -> BadIDChoice(14)
   slot 11 [flo,img] -> BadIDChoice(14)
   (both orders rejected here, although the previous turn recorded a REPLY for
    the same request shape -- that difference was not explained)

Effect on the photoflo: its p44 changed 00386510 -> 00382E10 and its band array
became 00382CD0 00386510 ... , i.e. the bind DID touch the photoflo even though
the request reported an error.  The image resource was unchanged.

NOT RESOLVED: what the element record's field layout is (its first word 0x01050001
does not decompose into a recognisable type/flag pair), what the bind is meant to
write, or how the element chain is meant to be constructed.  Without the 2.3
element encoding these dumps cannot be turned into a conforming pipeline.
