Fix bug in parsing of tasks in monitor, it was not using flatbuffers.

[robertnishihara]

This should address #354. When we updated some of the Redis payloads to be serialized using flatbuffers (in #341), we forgot to update monitor.py. This means that whenever RAY.TASK_TABLE_GET was called from cleanup_task_table in monitor.py, the monitor incorrectly parsed the local scheduler ID, and since the local scheduler ID was garbage, the monitor assumed that the task had been assigned to a dead node and so marked the task as lost. This caused the an attempt to reconstruct the task, which was fatal error because the task was an actor task.

So presumably the test failure was happening when the following conditions were met:

1. A task was already present in Redis when the monitor started up.
2. ray.get was called on the task triggering reconstruction.
3. The task was an actor task, since those currently can't be reconstructed.

[AmplabJenkins]

Merged build finished. Test PASSed.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/305/
Test PASSed.