The tmol How To guide notebook

[aleaverfay]

We will need to flesh out this notebook as we document for users how to do things with Tmol

[codecov]

Codecov Report

❌ Patch coverage is 74.38017% with 31 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.70%. Comparing base (a3e0864) to head (6b4c461).

Files with missing lines	Patch %	Lines
tmol/score/_score_function.py	65.43%	28 Missing ⚠️
tmol/relax/_fast_relax.py	70.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #399      +/-   ##
==========================================
- Coverage   90.74%   90.70%   -0.04%     
==========================================
  Files         445      445              
  Lines       36413    36463      +50     
==========================================
+ Hits        33042    33074      +32     
- Misses       3371     3389      +18     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[aleaverfay]

We will probably not use any part of the notebook that I've created, but I do have several changes to the API that I didn't want getting lost. Let's merge this as it stands and continue on with the documentation refinement in its own branch.

[kierandidi]

Audited against current master and #465. The important ScoreFunction on/off-term fix has already been reimplemented in current code with regression tests (including shared multi-score-type terms and weight-tensor refresh). The notebook was already identified here as disposable. The remaining API patch should not be merged as-is: cartesian_fast_relax selects default_kin_min_fn when no minimizer is supplied, PoseStack.to only changes coords despite kernels being intentionally float32-oriented, and the eager top-level imports undo the current lazy-import design. This branch is conflicting and its old green check does not validate current CPU/CUDA/PyTorch combinations. I recommend closing it as superseded; any still-wanted convenience API should be proposed separately with current end-to-end tests.