Need to be able to load custom models and lexicons from S3

[andrewresearch]

trafficstars

Perhaps offer optional S3 integration that is checked for on startup. Require a general bucket for tap set in an environment variable, then create standard directory structure for:

1. models - pre-trained models
2. lexicons - static lexicons
3. task - files related to specific tasks in uid directories
4. custom - everything else

[andrewresearch]

Use the batch mode #186 code to facilitate this process at startup.