Add regression coverage for large-project download flows

[dkritarth]

Problem

The existing tests cover basic project download behavior, but there is no explicit regression coverage for the large-project archive path or for keeping the CLI/MCP/git-remote flows aligned after the download implementation changes.

What should be covered

• Pulling a nested project structure to disk
• Downloading an archive file and verifying it is valid
• Preserving pull --force behavior and the "skip newer local files" safety check
• Verifying downloadByPath() fallback behavior still works
• Exercising the shared extraction path from MCP pull_project
• Exercising the shared archive path from git clone overleaf::...

Expected behavior

• A large or deeply nested archive fixture should be used to simulate the risky code path.
• The regression should catch memory-heavy or extraction-only failures before they reach users.
• Existing single-file download tests should remain intact.

Acceptance criteria

• Add a synthetic large archive fixture or equivalent integration test coverage.
• Test pull on a nested project tree.
• Test archive export parity between the canonical archive command and the compatibility alias.
• Test MCP pull on the shared extraction implementation.
• Test git remote clone/fetch using the same archive fixture or equivalent flow.

Notes

If a truly large archive fixture is too heavy for CI, a synthetic deep-tree archive is acceptable as long as it exercises the same extraction path and failure mode.

---

Generated by GPT-5.4 mini.