Draft: add in-repo Rust backend connector package

[gustavorubim]

Summary

This is a draft comparison PR for the architecture Miles asked to evaluate: the Python/Rust connector package lives in the PySR repository instead of in astroautomata/symbolic_regression.rs.

It is intentionally based on the existing Rust backend adapter PR (#1242), then adds an in-repo pysr_rust_backend/ subproject that builds the pysr-rust-backend distribution and exposes the symbolic_regression_rs import module consumed by PySR's adapter.

What this demonstrates

• Keeps the main PySR adapter plumbing from Add experimental Rust backend adapter #1242.
• Vendors the Python/Rust connector interface into this repository as pysr_rust_backend/.
• Leaves the Rust engine crates in astroautomata/symbolic_regression.rs; the connector uses Git dependencies for symbolic_regression and dynamic_expressions.
• Adds connector CI that installs the subpackage, runs wrapper tests, builds wheel/sdist artifacts, and smoke-tests those artifacts.

Why draft

This is for design evaluation, not an immediate merge candidate. It lets maintainers compare two structures:

• Split package: connector package lives in symbolic_regression.rs via Add Python wrapper package symbolic_regression.rs#47.
• In-repo connector: connector package lives here in PySR while the Rust engine remains in symbolic_regression.rs.

Validation

• cargo check --manifest-path pysr_rust_backend/Cargo.toml
• cargo fmt --check --manifest-path pysr_rust_backend/Cargo.toml
• python -m pytest pysr_rust_backend/tests -q: 12 passed
• python -m pysr test rust: 14 passed with the in-repo connector installed
• Parsed .github/workflows/CI_rust_backend_connector.yml with PyYAML

Related PRs

• Main PySR adapter PR: Add experimental Rust backend adapter #1242
• External wrapper-package PR: Add Python wrapper package symbolic_regression.rs#47

[gustavorubim]

Hi @MilesCranmer, following up on the in-repo connector comparison you requested. This PR demonstrates the connector living in PySR, while astroautomata/symbolic_regression.rs#47 contains the wrapper-first design used by #1242. The initial contract feedback has been addressed and both paths were validated. Which placement would you prefer us to consolidate around? Once decided, I can update the selected branch and mark the appropriate PR ready for review.

[MilesCranmer]

Hi @gustavorubim,
This will require significant effort from me to review, if I do decide to pursue this exact design, so I probably won't be able to get to this for a long time. My apologies.
Best,
Miles