Feat/issue 16 model training robustness

[jeffreyHoelzel]

Summary:

• training logic uses multi-seeded training for downstream statistical analysis and credibility
• for example, run 3 training/testing sessions at 10 epochs each using 11,22,33 as random number seeds, resulting in 3 models
• add new CLI args for model hyperparameter control

Closes: Issue #16