reconcile SLG/recursive solver's attempts to merge multiple results

[nikomatsakis]

PR #754 tweaked the behavior of recursive solver to do a better job merging multiple results (and in particular to recognize the special case of "always true"), but it did not make a comparable change to the SLG solver, leading to some divergence in results. We should fix that!