I've run these changes with js-ipfs-api against go-ipfs and since the expected hashes changed, the tests fail, they both should yield the same hashes for the same files

I checked to see if it could have been the change from raw to file (that could had happen in the other direction), but it seems that unixfs-engine is good with regards to that.

Nevermind, I hadn't realised that the test files were relatively small.
By the way, the difference with go-ipfs is that, for a file that is only one chunk, go-ipfs does not put a single link parent in front of it (going against what it does when recursing).
Tests against ipfs-inactive/js-ipfs-unixfs-engine#118 should be passing now.