fix: fall back to GitHub for LFS downloads

[Daniiiil1]

Contribution path

• Linked issue or discussion: lfs.dimensionalos.com 404 #2495

Problem

The configured lfs.dimensionalos.com endpoint can return 404 for objects that are still available from GitHub LFS. After retrying only the primary endpoint, get_data() raises and simulation startup fails. Users currently have to discover and persist a local lfs.url workaround themselves.

Solution

After the existing primary-endpoint retries are exhausted, retry the same object once through the repository’s GitHub LFS endpoint using a per-command git -c lfs.url=... override. This does not modify local or global Git configuration. If both endpoints fail, the exception reports both failures.

Tests cover a successful fallback, the exact non-persistent command, and the combined-error path.

How to Test

uv run pytest dimos/utils/test_data.py -k lfs_pull

I also verified the fallback endpoint against the known cafe.jpg.tar.gz LFS object with:

git -c lfs.url=https://github.com/dimensionalOS/dimos.git/info/lfs lfs fetch origin --include=data/.lfs/cafe.jpg.tar.gz

AI assistance

OpenAI Codex with GPT-5 assisted with codebase analysis, implementation, adversarial review, and test generation. The change is intentionally kept as a draft for final human review before requesting maintainer review.

Checklist

• I have read and approved the CLA.

Closes #2495