A pass to remove performance penalty in lurettetop (now the panalty in less than 1%).
The idea was to compute source info lasily.

Add nonreg tests for ldbg + crazy-rabbit, as well as missing files (oops).

Add online help msgs for the Ldbg_utils modules.

Better source code information. In particular, I've attached a
'Syntaxe.val_exp option' to each lxm:Lexeme.t in the
CoIdent.scope_stack type. The rationnale is to bez able to display
the whole expression associated to a lxm in ldbg (versus only the
lxm).

Add (preliminary) source code information !!

Among other things:

- split ldbg into ldbg and ldbg_utils.
- refine the event content.
- add support for online help.

Now I can trace lutin step called from run statements!

I've duplicated a lot of code in LutExe ; i'll fix that later...

I can now trace Lutin step.

The idea is to make lurettetop a caml toplevel that launches in
coroutine the whole test execution.

In order to implement that, I broke the inner loop in RunDirect.f
replace by a continuation that will make the coroutine easier.

Next step is to do the same in the lutin step.

nb : the gnuplot window needs to be refresed manually ('a' keystroke).
I should find a way to refresh it automatically.

It is a bit tricky because all the tools are compiled with the mingw
compiler, which means that all the system calls are done via the DOS
convention. Therefore, we need  to use :

- the rigth path conventions
- .bat instead of .sh scripts

Add a  --ocamlc-version option to lurettetop and lutin.

The idea is to put all the .exp files into a `lurettetop/lutin --ocamlc-version`
dir so that I have an .exp file for each ocaml version.

If the `lurettetop --ocaml-version`/*.exp does not exist, I generate it.

Also, don't non-reg test lucky dirs anymore (because they are boring).

Aldo, don't rely on the HOSTTYPE env var, but use Sys.os_type instead.

The lists of "true variables" and "true variables for the first time"
were interchanged.

Fix a bug with --delay-env-out option: the rif file was wrong (luciole outputs were appearinf twice).

returning wrongs values.  The problem was due to the use of caml
integer, that were not always big enough (on 32 bits arch). Bertrand
fixed by usings the Nums ocaml lib.

This breaks non-reg tests in lutin/test_ok because the exc Polka.overflow
is raised whereas the previous version was silently doing something wrong.

et "let x = i"

highligth the following bug: it is impossible to load from the same
main the same lutin program twice.  This occurs for instance with
lurettetop if 2 different main nodes are defined in the same file.

ZZZ : test4 should be re-activated (in the test rule) when this is
fixed.