We really should single thread ParallelFor in threadpool in the same way we single thead the schedule method

[tomasreimers]

https://github.com/tomasreimers/tensorflow-emscripten/blob/49456ba8ab874fd36edbf913745a2723494b39d1/tensorflow/core/lib/core/threadpool.cc#L125