For the price of loading transactions more consistently, we get a better subtitle. We do this in all cases EXCEPT for when we're grabbing handles, because that makes the handles pretty heavy weight and I could even feel the perf hit on my development machine and we don't use subtitle there anyway. We may want to cache the latest message on the conpherence thread object to improve performance here as well as consider falling back to "A, B, C..." more often. Code is written such that no transactions means an automagical fallback.
Fixes T7795. (Technically, there's still a note about handle code conversion work on T7795 but we'll get that generally later.)
