It's a bit more inconvenient in terms of register protection (BC
is much more generally useful than IY), but it makes tight spots
such as next and execute much faster, so I think it's worth it.

The 1 byte limitation has been effective for a while now, but I
hadn't made the move yet, I wanted to see if the limitation would
cause me problems. It doesn't.

Doing this now slightly facilitates the IY->BC move in z80.

Bootstrapping: if you try to recreate the CVM binary from the
previous commit with this code, you'll have bootstrapping problems.
The first bootstrap will compile a binary with 2-bytes wide cells
but branching conditionals that yields 1-byte cells. That's bad.

I got around the issue by temporarily inserting a "397 399 LOADR"
instruction in cvm/xcomp.fs, right before the xcomp overrides. This
way, I force 1-byte cells everywhere on the first compiliation,
which then allows me to apply the logic change in cvm/vm.c and have
a properly running binary.

For the IY->BC move, of course.

In preparation for the IY->BC move

Bring native words together, make the next/exec/does/compiled
cluster be at the beginning.

Next step: bring find code into _find word.

I'm trying to move stuff around and it simplifies things. Will make
a lot more sense to inline when BC replaces IY.

This is in preparation of using BC instead of IY for IP. I've
verified all PUSH* usage and none was problematic.

There is now no more actual code in stable ABI, only references.
This makes refactoring of this code much easier. For example,
changing IY to BC as the IP register.

Only its jump addr remains.

This allows us to refer to pushRS/popRS routines by labels, allowing
us to get rid of the pushRS reference in the stable ABI zone.

Only its jump at 0x33 remains.

I've also fixed a strange offset oddity in 8086's (n) placement.
It was off by 2, but strangely, it ran properly. Anyway, now it's
fixed.

Previously, it was impossible to cross-compile Collapse OS from a
binary-offsetted Collapse OS because stable ABI wordrefs would have
a wrongly offsetted address.

This solves the problem by replacing those wordrefs by direct,
hardcoded stable ABI offset references.

So far, I hadn't managed to run those tools properly on OpenBSD. I
was too confused by its stty peculiarities. I'm still confused, but
at least I managed to make them work... most of the time...

My intention is to reuse the SPI relay to program AVR MCUs from a
RC2014.

U didn't properly handle inserting on the last line and O's bounds
checks didn't make sense.

It's more descriptive this way. Also, I'll soon add a new HERESTART
config.

Driver configuration don't need their own words at runtime, we only
need to compile them as literals when compiling words.

Now that we have this "declaration blocks" pattern emerging, it
seems like a good idea to take advantage of this in drivers, both
for simplifying the xcomp unit and to make final binary slimmer.

Most of the time, this output would be hidden by the next screen
update, except when editing the last line.

Initially, I used the same letters as those used in the z80 ref
docs, but it makes the different assemblers harder to use than they
should. Having consistent "argtype" rules across assemblers should
help.

The stage binary is now there.

The C VM now runs the show.

Previously, recipes that began spitting binary contents before
loading block 282 would end up with VARIABLE code in their binary,
thus breaking them. We fix this by making this loading process
2-part.

Before lblnext is set, we need to use BJP, instead of JPnn, for the
jump to work in binaries with offsets.

Much faster!

This will soon replace the libz80 based "forth" and
"stage" executables. This is much, much faster.

I has removed it when I switched to Forth because it was emulating
the assembler version, but Michael updated it, so I relink it.

The JS VM part is so cool!