trunk/117872455ae8c1a04a03fd58498e0df6ec42e7f3: [BE] Do not checkout submodules during test jobs (#176655)

The PyTorch development team, maintaining the massively popular open-source machine learning framework with over 98,000 GitHub stars, has implemented a subtle yet highly effective optimization to their continuous integration (CI) workflow. In commit #176655, engineer 'malfet' applied a change suggested by contributor @yangw-dev, instructing test jobs to skip the automatic checkout of Git submodules. Since these submodules contain supplementary code not required for core testing operations, their exclusion represents pure overhead. The result was an immediate and dramatic speed-up, transforming a previously 75-second procedural step into a 15-second operation. This 60-second saving per test job compounds significantly across the project's vast and active development pipeline.

This optimization matters because PyTorch's CI system handles thousands of test runs daily from both internal Meta developers and external contributors. Reducing a bottleneck by 80% directly accelerates the feedback loop for engineers, allowing faster iteration on pull requests and quicker identification of regressions. For an ecosystem-critical project foundational to AI research and production, such incremental efficiency gains are vital for maintaining velocity. The change highlights how mature open-source projects can achieve substantial performance improvements not just in runtime code, but in the developer tooling and infrastructure that supports it, ensuring the framework remains agile as it scales.