CLOSURE: no client request writes object->byte8.  The transition into the
importable state is unreachable from the request set.

Method: valid c.new_id ids, no-reply model (silence = success), generated bit
addresses.  A single breakpoint captured the object pointer from create, then a
WRITE WATCHPOINT was installed on byte 8 of that object and left armed across a
complete cycle.

  create  -> silent (success)      obj = 003FDC50
  watch installed at 003FDC90      (= obj + 8 bytes), byte8 read-back = 1
  arm1    -> silent (success)
  put     -> BadMatch
  arm2    -> BadMatch
  export  -> silent (success)

  byte8 WRITE EVENTS AFTER CREATE: 0
  byte8 value throughout          : 1

So nothing in create-after, arm, import, re-arm or export writes byte8.  Combined
with the code findings:

  arm    requires obj->byte8 == 1   (slot 20, file 0x6ACBE/0x6ACC2)
  import requires obj->byte8 == 3   (slot 19, file 0x6BAF4/0x6BAF6)
  create sets byte8 = word9 >> 8    (file 0x6CB0A/0x6CB12)

the only way to obtain byte8 == 3 is create with word9 = 0x300, and such a
resource cannot be armed (arm returns BadMatch at 0x6ACD4 for every one of the 25
(v12,v14) combinations tried).  An unarmed resource's import returns BadAccess
without reaching the gate.

CONCLUSION (with evidence, not assertion): a byte8 transition into the importable
state is NOT reachable from the client request set.  Therefore the import cannot
be driven, src_base remains 0, and EXPORT has nothing to draw.  This closes the
last request-level door identified in this investigation.

Residual possibilities, all outside the client request set:
  * an internal code path that writes byte8 (the watch shows none occurs in a
    normal cycle, but a path we cannot trigger could exist);
  * a request form with a shape never constructed (every implemented slot has
    been swept, so this would have to be an unimplemented minor);
  * the DEC XIE 2.3 protocol may define a different intended sequence entirely,
    which the documentation would state directly.
