Tasks which handle datasets should be run in separate process

[skatedplrn]

if task failed with OOM , hub does not get any result, since we do not use result backend see trial worker in auger-ml