fix: avoid torch mirror as primary index

[benceruleanlu]

Context

• macOS fresh installs consistently failed during the compiled requirements step and the torch fallback step when download.pytorch.org timed out (e.g. resolving pillow as a transitive dependency with --index-url pointing at the torch mirror).
• The macOS compiled lockfile included a PyTorch extra index, so uv could resolve non‑torch deps from the torch index, which made the install brittle when that host was slow or blocked.

Changes

• Remove the PyTorch extra index from assets/requirements/macos.compiled to keep macOS compiled installs on PyPI only.
• Keep PyPI as the primary index for torch installs and treat PyTorch mirrors as extra indexes, while still allowing custom mirrors to remain primary.
• Recognize torch mirrors via download.pytorch.org/whl/(cu*/nightly) patterns so older CUDA indexes are supported.
• Add unit coverage for the torch mirror detection behavior.

┆Issue is synchronized with this Notion page by Unito

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}