Currently flush_tier iterates through all tiered: tables in the metadata and attempts to flush each of them.  This is extra work for tables that haven't changed since the previous call to flush_tier. 

We should implement something (similar to checkpoint) that only flushes tables that need to be flushed.  Note that a table that needs to be flushed might not have any dirty data in the cache and may not have an open dhandle. I.e., we might have checkpoints and closed the table since the last flush, but the data from that checkpoint still needs to be flushed.