In txn_global_query_timestamp, when retrieving the all_durable timestamp, we walk the array of concurrent transactions, search for the minimum, and return a value one lower than that minimum.

However, in the case where either the first transaction has the lowest durable timestamp, or there's only one transaction with a durable timestamp, we'd set `*tsp` one too high. I believe the code should look more like this:

        /* Walk the array of concurrent transactions. */
        WT_ORDERED_READ(session_cnt, conn->session_cnt);
        for (i = 0, s = txn_global->txn_shared_list; i < session_cnt; i++, s++) {
            __txn_get_durable_timestamp(s, &tmpts);
            if (tmpts != 0 && (ts == 0 || tmpts < ts))
                ts = tmpts;
        }
        if (ts != 0)
            --ts;

Also in scope: adding a test, and verifying the supposition that this should return a timestamp "one lower" than the lowest durable timestamp at all.