Time elapsed msg does not account for time before any preemptions

> @dustinvtran
> @ywen666
>
> We should save the time elapsed number in checkpoints. This enables pre-empted jobs to add on top of the number instead of restarting from 0 but with a later checkpoint.
>
> Looking at https://www.tensorflow.org/api_docs/python/tf/train/Checkpoint, this requires storing `start_time` as `tf.Variable(start_time)`.

https://github.com/google/edward2/issues/260