Run threaded tests on empty cluster

so that it is guaranteed that lazy workers can be added. Also, the tests
make sure to remove all workers that had been added, so that subsequent
tests can determine the laziness of the cluster themselves.

stdlib/Distributed/test/distributed_exec.jl

@@ -1709,9 +1709,9 @@ for p in procs()
 end
 
 include("splitrange.jl")
-include("threads.jl")
 
 # Run topology tests last after removing all workers, since a given
 # cluster at any time only supports a single topology.
 rmprocs(workers())
+include("threads.jl")
 include("topology.jl")