ACCELERATOR MODEL, ITERATION 2 -- THREE WAITS CLEARED, NOW BLOCKED ON THE FIFO
HANDSHAKE (which must be stateful, not a constant).

Method: with VT1200_ACCEL=1 plus the labelled compression poke, sample the guest
PC during the freeze (emu.register_periodic) and map each sample to a file
offset.  Each wait cleared moved the spin to the next one.

  1. 0x122E54  polls bit 1 of 0x5600010  -> "initialisation of scalution"
       model: return bit 1.                                          CLEARED
  2. 0x1253FC  polls bit 9 (done) or bit 0 (data ready) of *B5, error mask 0x450
       model: add bit 9.                                             CLEARED
  3. 0x125448  polls bit 2 or (bit 9 of *B3)
       model: add bit 2.                                             CLEARED
  4. 0x1254F2  polls bit 0 (data ready) -> transfer path
       model: add bit 0.                                             REACHED
  5. 0x125568..0x125570  write loop: MOVI 0x7FE,B8 ; MOVE A11,*A5 ; DSJS
       -- writes 0x7FE words to the FIFO data register, then returns to
          0x1254F8 -> 0x1254F2, which sees bit 0 still set, and re-enters.
       => infinite transfer loop; the guest freezes here.

So the model needs the DATA-READY bit to be *transient*: asserted so the engine
takes a transfer, deasserted once the FIFO write happens (and re-asserted when
the engine wants more), rather than held permanently as it is now.

Current status value in the model: 0x0207 (bits 0,1,2,9) -- one bit too many:
bit 0 must be a handshake, not a constant.

STATE
  * default configuration unchanged and verified booting (model is opt-in).
  * with VT1200_ACCEL=1 + poke the export advances through three accelerator
    gates instead of failing at the first, and now blocks in the FIFO transfer.
  * NO client-chosen XIE image has been displayed; the window is unchanged;
    xshow (core X) remains the only viewer that has shown an image.

NEXT: make 0x5600010 bit 0 a latch -- set when the engine writes the FIFO command
(0x5600080 / 0x5600050) and cleared by the first FIFO data write -- i.e. give the
accelerator a one-word-deep "data request" handshake.  That is a small,
well-scoped change to accel_r/accel_w in harness/mame/vt1200.cpp.
