Support different temperatures in batches

[orionarcher]

Summary

Include a summary of major changes in bullet points:

• allow different temperatures to be specified in batches in NVT, issue Allow batched integrators to evolve systems at different temperatures #44
• remove batched initialize momenta because it is deprecated and required all structures to be the same length
• add tests for calculate momenta to ensure momenta removal works
• add warning if sevennet is initialized with float64 Floating Point Exceptions with Sevennet #117
• refactor to prevent optimize from calling init_fn on all states at once [Enhancement] Initial evalulation of state causes an OOM error if state is too large #128

Checklist

Work-in-progress pull requests are encouraged, but please enable the draft status on your PR.

Before a pull request can be merged, the following items must be checked:

• Doc strings have been added in the Numpy docstring format.
  Run ruff on your code.
• Tests have been added for any new functionality or bug fixes.
• All linting and tests pass.

Note that the CI system will run all the above checks. But it will be much more
efficient if you already fix most errors prior to submitting the PR. It is highly
recommended that you use the pre-commit hook provided in the repository. Simply run
pre-commit install and a check will be run prior to allowing commits.

[abhijeetgangan]

LGTM. Nice use of scatter operation, btw :)

[abhijeetgangan]

Something seems wrong in the graph pes test- https://github.com/Radical-AI/torch-sim/actions/runs/14406977027/job/40405760127?pr=123

Unrelated to this PR.

[orionarcher]

@jla-gardner @CompRhys, seems like one of the graph-pes tests fails intermittently, I am inclined to just loosen the bound to 5e-5?

[CompRhys]

@jla-gardner @CompRhys, seems like one of the graph-pes tests fails intermittently, I am inclined to just loosen the bound to 5e-5?

I realize that I didn't implement what I thought I had for the rattle, I was planning to do closer to

    perturbed = struct.copy()
    for site in perturbed:
        magnitude = rng.weibull(gamma)
        vec = rng.normal(3)  # TODO maybe make func recursive to deal with 0-vector
        vec /= np.linalg.norm(vec)  # unit vector
        site.coords += vec * magnitude
        site.to_unit_cell(in_place=True)

based on CGCNN+P from MBD but didn't as I didn't sample the directions. We should just replace with a gaussian rattle will be easier.

--edit

I fixed this in 1344660.
The Weibull gives a much nicer distribution for rattling so I stuck with it rather than going Gaussian.