LCG update

[cprudhom]

Some improvements to the LCG framework:

• MiniSat now manages new WL in a circular way (like depicted in "Optimal Implementation of Watched Literals and More General Techniques", Ian P. Gent, JAIR 2013)
• When a clause is learned, the 2nd watched-literal is the one with the highest decision level (as defined in "An Extensible SAT-solver", Niklas Eén, Niklas Sörensson)
• LCG now call for DB reduction before adding the last nogood
• Profie (for Cumulative) lazily computes the set of tasks involved in a conflict

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request

[ArthurGodet]

The code changes seem fine to me. The only remaining thing to dois change the number of expected nodes in tests in choco-parsers module.

Maybe we should consider removing these specific asserts (and keep only the ones of the objective value or number of solutions). What do you think ? @jgFages an opinion on the matter ?

[cprudhom]

In my opinion, we should keep these metrics. It is admittedly a bit of a pain to update them whenever we make a change that we know will affect them, but it also helps us spot regressions caused by other, indirect changes (such as, for example, when we change certain constraints).

[jgFages]

I would tend to agree with both :-)
It is good to keep (at least some of) them in order to detect potential regressions.
It is also acceptable to remove some node/fail counts, especially when using dynamic strategies.