Use CPU/GPU extras with uv sources for torch installation

[RajdeepKushwaha5]

Describe your changes

Move torch from direct [project] dependencies to [project.optional-dependencies] with mutually exclusive cpu and gpu extras. Add [tool.uv.sources], [[tool.uv.index]], and [tool.uv] conflicts so that uv sync --extra cpu (or --extra gpu) installs torch from the correct PyTorch index in a single command.

Simplify the uv CI path from manual uv venv + uv pip install torch + uv pip install -e . to uv sync --extra ${{ matrix.device }} --group dev. The pip CI path retains the dry-run hack since pip cannot read [tool.uv.sources]. README install instructions updated accordingly.

Issue Link

closes #600

Type of change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)
• 📖 Documentation (Addition or improvements to documentation)

Checklist before requesting a review

• My branch is up-to-date with the target branch - if not update your fork with the changes from the target branch (use pull with --rebase option if possible).
• I have performed a self-review of my code
• For any new/modified functions/classes I have added docstrings that clearly describe its purpose, expected inputs and returned values
• I have placed in-line comments to clarify the intent of any hard-to-understand passages of my code
• I have updated the README to cover introduced code changes
• I have added tests that prove my fix is effective or that my feature works
• I have given the PR a name that clearly describes the change, written in imperative form (context).
• I have requested a reviewer and an assignee (assignee is responsible for merging). This applies only if you have write access to the repo, otherwise feel free to tag a maintainer to add a reviewer and assignee.

Checklist for reviewers

Each PR comes with its own improvements and flaws. The reviewer should check the following:

• the code is readable
• the code is well tested
• the code is documented (including return types and parameters)
• the code is easy to maintain

Author checklist after completed review

• I have added a line to the CHANGELOG describing this change, in a section
  reflecting type of change (add section where missing):
  • maintenance: added entry under Maintenance in CHANGELOG.md

[Copilot]

Pull request overview

This PR restructures how PyTorch (torch) is installed by introducing mutually exclusive cpu/gpu extras and using uv source routing to automatically select the correct PyTorch index, simplifying both CI and contributor installation flows.

Changes:

• Move torch from base dependencies to cpu/gpu optional extras and add uv index/source routing plus extra conflicts.
• Simplify the uv CI installation path to uv sync --extra ${{ matrix.device }} --group dev.
• Update README installation guidance and add a Maintenance entry to the changelog.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
pyproject.toml	Adds cpu/gpu extras for torch and configures uv indexes/sources with conflict rules.
README.md	Updates installation instructions to use uv sync --extra cpu/gpu and clarifies pip’s manual torch install.
CHANGELOG.md	Documents the maintenance change for torch installation approach.
.github/workflows/install-and-test.yml	Updates CI to use uv sync for uv installs and adjusts the pip dry-run logic.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[sadamov]

This looks good overall. The extras + [tool.uv.sources] approach is the right solution for uv sync. A few things to fix:

This PR has a big! side-effect, changing the cuda version from 11.1 to 12.8 during install instructions. This should be made very clear to everyone and also discussed at a future dev-meeting.

[sadamov]

Nit: name : has a spurious space (line 46 of the updated workflow file). Since this line is between diff hunks it cannot be an inline suggestion, but it should be fixed:

      - name: Install uv (if applicable)

[RajdeepKushwaha5]

Pushed fb8f4eb addressing all review comments. Summary of changes:

pyproject.toml

• Restored torch>=2.3.0 to [project.dependencies] so pip install neural-lam (and uv sync without an extra) keeps torch as a hard requirement. The cpu/gpu extras now only select the variant.
• Added the --torch-backend / uv issue #12994 note above [tool.uv].

.github/workflows/install-and-test.yml

• Rewrote the header comment per your suggestion (torch framed as required, extras select variant, --torch-backend note, pip routing explanation, hosts block preserved).
• Reverted the pip dry-run argument from ".[${{ matrix.device }}]" back to "." — pip ignores [tool.uv.sources] and torch is in core deps now, so the extra is unnecessary.
• Fixed the name : whitespace on the Install uv step.

README.md

• Replaced the **regarding torch installation** paragraph with your shorter wording that makes the hard requirement explicit.
• Expanded the uv install steps with the post-install uv pip install torch --torch-backend auto recipe for users on a CUDA version other than 12.8.

Commit message

• Updated to reflect that torch stays in core deps (the previous wording said "Move torch from direct dependencies to optional-dependencies" which is no longer accurate).

[sadamov]

Thanks for the round of revisions in fb8f4eb, all my previous comments are addressed. Three things left, and I'd like all of them in this PR rather than as follow-ups so we land the lockfile + cache-key story coherently with #606:

1. CHANGELOG: link the PR, split into two entries to surface the CUDA bump.
2. Commit uv.lock and switch CI to --locked. This is the big one.
3. README: lead the CUDA-swap recipe with --extra cpu (smaller download) and clarify the swap is reverted by the next uv sync.

Details inline.

[sadamov]

This looks ready! the bug fixed in #639 remains fixed. Lock-file adds reproducibility layer and also allows for easier caching and faster CICD.
Thanks a lot @RajdeepKushwaha5!

[joeloskarsson]

All looks good to me! Now I have not been involved with the original gh workflow setup, nor am particularly knowledgeable about it, so leaving it to you if you want to get an additional review from someone who knows it more. But I understand all these changes and think they are good.

[observingClouds]

LGTM, just the AGENTS.md needs an update about the new install instructions.

[leifdenby]

Just had a read through and this looks great to me too 🌟 Thank you for implementing this @RajdeepKushwaha5! 🚀

[observingClouds]

LGTM, thanks for addressing my comments.

[RajdeepKushwaha5]

thank you everyone for the detailed reviews and patience. I learned a lot through this PR, especially around uv, PyTorch installation flows, lockfiles, and CI reproducibility. I’m already working with CGAL for GSoC this year, but I’d genuinely love to stay involved with MLLAM as well if there’s any way I can continue contributing or help in some role outside GSoC.

Really appreciate the guidance from everyone here!