I believe these tests are incorrect, and expect subgrid margin/border/padding contributions to be capped to the size of the edge track. https://github.com/w3c/csswg-drafts/issues/7072 is open to discuss this, and figure out a resolution.
<rdar://problem/89218999>
The CSSWG issue has been resolved, and these tests are definitely invalid. The remaining work here is rewrite the tests to be correct, though it might be simpler to start from scratch and write new tests for subgrid margin/border/padding being larger than the edge tracks.