Add driver ID to task spec and add driver ID to Python error handling.

[robertnishihara]

This provides a small amount of isolation between different drivers using the same cluster (there is not much isolation at the moment, e.g., remote functions, environment variables, and such will conflict).

The following can be done to cause an error from one driver to be pushed to another driver.

From ray/ start a local cluster with

./scripts/start_ray.sh --head

Then in two separate windows, attach drivers to the cluster by following the instructions printed out from the start_ray.sh command, e.g.,

import ray
ray.init(redis_address="...")

Then define a remote function on one of the drivers that throws an exception, and call it.

@ray.remote
def f():
    raise Exception("ERROR!!")

f.remote()

The error message should appear in both windows.

[pcmoritz]

above, define a global variable ERROR_KEY = b"Error:" (we have a convention for naming here I think and also if we include ":" or not) and replace 5 + 1 by len(ERROR_KEY) and everywhere else replace b"Error:" by the variable

[pcmoritz]

same for "6" here