Bug fixes for Proxima implementation

[WardLT]

Caught a bunch of things while testing our on-the-fly learning implementation

• Making a TorchANI calculator from a model sets it as not-trainable
• Loading the Proxima state does not update the model parameter

Also added some improvements to the testing script

• Start a run by reading a parameter file from disk
• Ensure a minimum fraction of steps are performed with DFT
• Write training logs to disk (this is what helped me catch the above bugs)

[coveralls]

Pull Request Test Coverage Report for Build 9940363502

Warning: This coverage report may be inaccurate.

This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

• For more information on this, see Tracking coverage changes with pull request builds.
• To avoid this issue with future PRs, see these Recommended CI Configurations.
• For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

• 16 of 16 (100.0%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.1%) to 94.386%

---

Totals
Change from base Build 9910390088:	0.1%
Covered Lines:	538
Relevant Lines:	570

---

💛 - Coveralls