feat: cosine_similarity function

[universalmind303]

adds new function cosine_similarity for comparing similarity across floating point vectors. cosine_similarity is commonly used in vector search workloads.

[greyscaled]

I tried it out, and it worked nicely.

I don't fully understand why two equivalent vectors gives me 5.9604645e-8 instead of 0.0 (as shown in the tests), but I understand that this is upstream behaviour and I lack some prior art/experience.

[universalmind303]

I tried it out, and it worked nicely.

I don't fully understand why two equivalent vectors gives me 5.9604645e-8 instead of 0.0 (as shown in the tests), but I understand that this is upstream behaviour and I lack some prior art/experience.

yeah i think this is just a math/precision thing when performing similarity. 0.000000059604645 is practically 0.0

[universalmind303]

@tychoish the CI seems to keep getting stuck on build cache. I don't think this is related to this PR at all.

https://github.com/GlareDB/glaredb/actions/runs/8008291239/job/21874514119?pr=2680