Refactor IRJ parser for Menhir new error handling

[mdurero]

As stated in #238, traditionnal error handling is deprecated in favor of a flexible approach using Menhir incremental API.

Close #238

[mdurero]

The current architecture is kind of underwhelming.

On the one hand I would find satisfying to merge the parse_file and dummy_parse_file_with_incremental in a unique parser entry point, to reuse the lexbuf for instance.
On the other hand, in Mlang interpreter context and other "not validation" ones, do we want the full two pass parser to be called when the first pass would be enough ?

Yes, writing it I just realise that the second pass is called only on errors, so it doesn’t slow the interpreter in a right case. But I keep the question as my problem is more than both pass have a very distinct behaviour. The second pass prints on error output in all cases.

So at the end, my proposition would be:

• modify second pass to create a proper error type with its message, but do nothing with it,
• the irj_checker displays all errors,
• other context like the interpreter can choose to not do that and fail silently, or be less verbose.

At the end, I feel like irj_checker could also almost be just a wrapper calling parse_file with a verbose flag…

[mdurero]

I think we should slow the work on the json generator and focus on the polish on file validation to merge it soon.

[denismerigoux]

OK, let's merge!