it looks good to me but i have a couple of thoughts:

1 - i know you put effort into keeping the git history but i'd actually squash this anyway if we merge it as i don't think the subset history makes sense in our tree. it may be better to just note somewhere this was originally built by andrii

2 - something's bugging me about the overlap here of includes. we have basically the same algorithm in includes (written differently) than we do here, but in that one we don't do partial equality on the deep properties.

i wonder if we can at least extract the traversal logic so they both share it, or we make includes have a flag that achieves what subset does. im cautious of putting too much logic in includes though as its already a pretty big, multi-use assertion.

2 - something's bugging me about the overlap here of includes.

We'll likely have similar logic in deep-eql too. I'll add that deep-eql is also very well optimised. I wonder if we can leverage it and it's custom comparators to effectively traverse the first layer?

I'll start looking at squaring this up according to the comments.