What is the rationale of this change? If the futures_factory was created with an executor, this executor should be used to schedule the task. If we applied this patch, then

 pool_executor exec{"foo"};
 auto f = async(exec, ...)

would schedule the task on the default pool instead of on the requested one.

As an alternative (to make things more consistent) we could either rewrite the implementation of async for thread-executors (like pool_executor, see https://github.com/STEllAR-GROUP/hpx/blob/master/hpx/parallel/executors/thread_execution.hpp#L47-L57) or we could change future_factory to always use an executor to schedule things, even if it was not created with one. The second however, would probably complicate things even further.

I am closing this issue because it turns out to be a red-herring. My work on #3241 was in effect calling some async overloads twice and after fixing that, I do not need this PR any more.