Bug fixed in two lines + 5 lines of comment.

Bug fixed in two lines + 5 lines of comment.
During select, the context of the goal is changed and the conclusion
is changed too (too a ?[out_scope]). Then the changed conclusion was
relocated in the changed context, triggering a delift that in turn triggered
the unification case "something vs ?[out_of_scope]"... at the bad time!
Fixed by manually relocating the new conclusion in the new context.
Manually means opening and building again the cic_term data type, which is
ugly and sort of fragile.