[bugfix] Avoid forgetting 1 atom clauses

When a new clause with only 1 atom is added with (usually
from simplifying a theory lemma), the usual strategy is to backtrack to
the base level, then propagate the atom. However, when the atom is
already false at base level, Unsat is raised, in which case the
information that the atom must be true is lost. To avoid that, the
single atom clause is simply pushed back in the stack of clauses to add,
so that it will be re-examined later.

src/core/internal.ml

@@ -753,8 +753,8 @@ module Make
     Log.debugf 90 "Adding clause:@[<hov>%a@]" (fun k -> k St.pp_clause init);
     let vec = match init.cpremise with
       | Hyp -> env.clauses_hyps
-      | Lemma _ -> env.clauses_learnt
-      | Local | History _ -> assert false
+      | Local -> env.clauses_temp
+      | History _ | Lemma _ -> env.clauses_learnt
     in
     try
       let atoms, history = partition init.atoms in
@@ -776,6 +776,13 @@ module Make
         Log.debugf 5 "New unit clause, propagating : %a" (fun k->k St.pp_atom a);
         cancel_until (base_level ());
         if a.neg.is_true then begin
+          (* Since we cannot propagate the atom [a], in order to not lose
+             the information that [a] must be true, we add clause to the list
+             of clauses to add, so that it will be e-examined later.
+             Also, in order to avoid it being twice in a clause vector, we pop
+             the vector where it was added. *)
+          Stack.push clause env.clauses_to_add;
+          Vec.pop vec;
           report_unsat clause
         end else
           enqueue_bool a 0 (Bcp clause)