byte8 WRITE WATCH, NOW WITH A PROVABLE POSITIVE CONTROL.

Previous attempts at this watch had no control that could be shown to fire, so the
"0 writes" result was not trustworthy (and one VRAM control also read 0, which
told us nothing).

This run installs, alongside the byte8 watch, a control watchpoint on the STACK
region just below SP -- any function call pushes there, so it must fire if
watchpoints work at all:

  installed: byte8 watch at 003FDA50 ; stack control at 000052A0 (sp=000053A0)

  arm1      -> SILENT(ok)      stack control hits: 1218
  PUT_TILE  -> ERROR 8
  arm2      -> ERROR 8

  TOTALS:  stack-control hits: 1820      byte8 hits: 0

So:
  * write watchpoints DO work in this MAME build (1820 positive hits);
  * NOTHING writes object->byte8 during create -> arm1 -> PUT_TILE -> arm2
    (0 hits against that control).

This CONFIRMS, on validated instrumentation, the earlier claim that only create
sets byte8 (via word9 >> 8), and therefore confirms the mutual exclusion:

    arm (slot 20)                 requires byte8 == 1
    import gate (0x6BAF4)         requires byte8 == 3
    only create can set byte8

Together with the previous turn's validated gate trace (T_ENTRY -> T_DISP -> T_C2
-> T_J1 -> T_B32 -> T_MERGE -> T_GATE -> a8=8), the model is now:

  byte8 = 1 : arm succeeds; PUT_TILE reaches 0x6BAF4 and fails there (BadMatch)
  byte8 = 3 : arm fails; PUT_TILE never reaches the gate

and no client request can move byte8 between those values.  Within the request
surface I can construct, the two requests are mutually exclusive.

Status: no client-chosen XIE image; src_base 0; window still the stipple;
xshow (core X) is the only verified viewer.
