Test and fix bad positions in conflict message

[AltGr]

Related to #755, which this fixes for the lcalc interpreter ;
however, the real issue is the OCaml (and other) backends, where it's much harder.

Indeed, for those, the positions are passed as a flat list to the top-level default term, which has no information anymore about nested exceptions. The only reliable way I see to fix that is to switch evaluated default terms from 'a option to ('a * pos) option, and remove the the flat list of positions. It requires a change in the type of the handle_exceptions primitive, but it shouldn't be too difficult now that positions can be reified.

[AltGr]

(note that, although the test triggers it with the merged same-output exceptions, the issue would exist with any nested exceptions)

[denismerigoux]

I agree with this fix as well as the proposed generalized solution to change the accumulator type from 'a option to ('a, pos) option in the OCaml backend (and runtime.ml subsequently).

[denismerigoux]

Can I merge this or do you want to put the refactoring of the runtime positions in default in this PR?

[AltGr]

Hm, since ① this doesn't solve the issue where it is problematic (the backends) but only in the lcalc interpreter which the users shouldn't care about and ② the "real" fix should supersede this anyway, I don't think it's worth merging (although I'd like to keep it open for the testcase that is added)