rusttorch 0.3.0

Build, train, and run neural networks in Rust with LibTorch
Documentation
## Summary

<!-- What changed, why, and what is intentionally out of scope? -->

Closes #

## Validation

<!-- List exact commands, results, hardware backends, and intentional skips. -->

## Checklist

- [ ] The change is focused and an issue approved any large API, architecture,
      dependency, unsafe-code, or compatibility work.
- [ ] Tests fail without the behavior and pass with it; unavailable hardware is
      reported as skipped.
- [ ] Formatting, checks, Clippy, tests, rustdoc, and applicable package checks
      pass.
- [ ] Public documentation, examples, and `CHANGELOG.md` are updated when
      behavior changes.
- [ ] The compatibility ledger and generated coverage are updated for public
      compatibility changes, with PyTorch source and parity evidence.
- [ ] Performance claims include a reproducible benchmark, named baseline,
      environment, methodology, raw results, and variance.
- [ ] New dependencies and unsafe code have explicit maintenance, license, and
      security review.
- [ ] Third-party and generated material has documented provenance and a
      compatible license; notices are updated where required.
- [ ] No credentials, personal data, downloaded runtimes, generated model
      artifacts, or unrelated build output are included.
- [ ] I reviewed any AI-assisted output and remain responsible for its
      correctness, provenance, licensing, and security.
- [ ] I will follow the Code of Conduct and support policy.
- [ ] Every commit has a matching `Signed-off-by` trailer certifying DCO 1.1.