0.3.2

Improvements

• expose created_at and updated_at columns of tasks table via REST API
• separate model training and prediction for classifier tools to reduce memory usage and allow parallelization of label prediction
• speed up insertion of tool results into database by targeting individual shards directly
• no longer serialize type of workflow step arguments and enforce types rather than performing type checks
• implement ORM interface for distributed tables to allow modification of distributed tables via ExperimentSession (this further abstracts table distribution; users can now handle distributed tables the same way they would handle any other table)