Replace its usages with direct RAM+ offsets. The (sysv) mechanism was
incompatible with cross-compilation of a full interpreter.

This makes IMMEDIATE affected by memory mappings.

It copied one too much.

Doing this was a bit stupid. These offsets are constants. Moreover,
having them in stable ABI had us construct the boot binary from the
stable ABI of the host, making it very difficult to change RAMSTART
for a new system.

Not needed anymore. CONSTANT does the trick.

We refer to stable offset as direct numbers instead of offset to JTBL.
Simpler that way.

There is an alternate git history where I continued the Forth-ification of
words, including "SKIP?", but that was a bad idea: because that word was
written by flow control immediates, I stepped into quicksands where stability
became necessary in z80c.fs and I couldn't gracefully get out of it.

I'm stepping back and take this opportunity to replace the shoddy SKIP? algo
with a more straightforward (?br) implementation.

(br) and (?br) will always stay in boot code where it's easier manage a stable
ABI.

I can't get rid of "(fbr)" and "(bbr)" just yet, but soon...

Otherwise, when a defining word would be called outside a definition
itself, it would get the name of the last parsed word, that is,
itself!

For example, dummy.fs, instead of creating a "_______" entry, created
a "(entry)" entry...

The commit ended up being much bigger than anticipated. This was a long thread
of underlying complexities. This lead to the creation of interesting concepts
such as (sysv).

It's not really an App.

Also, un-IMMEDIATE "LITN" it didn't make any sense.

Previous to this commit, DEFINE's algo would miss a ";" right after a newline.

This allows for a much nicer formatting+comment of core.fs.

Change the mainloop so that words are executed immediately after they're read.
This greatly simplifies execution model and allow the "DEFINE" word to become
an IMMEDIATE and stop its "copy from compiled words" scheme.

The downside to this is that flow control words no longer work when being used
directly in the input buffer. They only work as part of a definition.

It also broke "RECURSE", but I've replaced it with "BEGIN" and "AGAIN".

Another effect of this change is that definitions can now span multiple lines.

All in all, it feels good to get rid of that COMPBUF...

Readline, instead of being triggered by the end of execution of the last
compiled line is now triggered "just in time", by "WORD".

This allows IMMEDIATE words reading input buffer to span multiple lines
( comments for example, but colon definitions will soon follow ).

This will allow us to support backward branching with just one new (bbr) word.
Also, this allow us to have "(" word sooned in core.fth and thus allow for
earlier commenting.

Also, fix "RECURSE" in IMMEDIATE contexts.