Spider: Use GROUP BY handler only when necessary and safe

In my opinion, Spider should not use GROUP BY handler unless it is necessary, for performance, and safe. That is because the GROUP BY handler reassembles the query from a general (possibly complex) execution plan, it tends to be more prone to mistakes than execution through the normal handler interface (which should result in a simpler query rassemblement). This is supported by the fact that many of the recently reported bugs are not reproducible on 10.2. I know that there are some problems with access via the handler interface (e.g., MDEV-26111), but I think it could be fixed.