Optimization for event_handler-in-Redis

[alanking]

Based on the work in #102...

Possible future optimization:

• Write down the target path to the custom event handler (which includes a uuid for purposes of idempotency) in the meta map
• If the worker sees that the file with that target path already exists, it will not need to go to redis to get the file contents
• If the target path does not exist, the worker will go to Redis to get the custom event handler contents to write down as it does in the work of #102 (using the event_handler_key)

This should reduce network traffic by N files - M num_workers at the cost of N files * target_path_length_in_bytes in Redis.

I think an option should be provided via the CLI to skip checking Redis entirely if the user knows what path the workers are supposed to be checking (i.e. restores the option of using the historical, "manual" deployment).

[trel]

this should prove efficient because network traffic is a lot more expensive than a larger redis database.

but we'll test this assumption before merging.

with N = 1 million files, M = 50 workers, and target path length = 40bytes... ... estimated savings of 999,950 redis hits for the cost of 40MB of additional redis memory