Support deploying external TensorFlow models

[deliahu]

Description

Allow users to deploy already-trained models as APIs

Notes

• Should cortex ingest the model into the cortex bucket, or read it directly from the external source in the API pods when they start up?