Add option to begin training from initial models

[WardLT]

I'm concerned that we're deviating too far from the original weights of a pre-trained model, so am adding an option where we finetune from the original model before training.

Fixes #50

To do:

• [x] Add to the proxima test script
• [x] Verify the test script works, comment on effect

[coveralls]

Pull Request Test Coverage Report for Build 10581941498

Details

• 9 of 10 (90.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.1%) to 95.172%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
cascade/proxima/init.py	9	10	90.0%
<!--	Total:	9	10	90.0%	-->

Totals
Change from base Build 10391705242:	-0.1%
Covered Lines:	749
Relevant Lines:	787

---

💛 - Coveralls