Be more cautious with model-related artifacts

[lgessler]

The problem: There are many places where we used serialized representations of parameters or other model-related information.

Why it's a problem: Often these are stored in a binary format (e.g. a pickle) that may be volatile with respect to platform, Python version, or dependency version.

Proposed solution: At minimum, these kinds of serialized representations should make explicit the assumptions they're making about platform and versions. This could mean:

• (prerequisite) explicitly indicate python and package versions for any given release of the AMALGUM pipeline
  • I'll be doing this soon with a conda env.yml file
• Adding AMALGUM version numbers to serialized model files stored on corpling.uis.georgetown.edu