batched cosine / euclidean procedure for more efficient computation of vector similarities

[jexp]

apoc.ml/algo.cosine(Similarity) or sth like that.

Parameters:

• list of nodes
• property name
• target embedding
• top-k
• threshold

Efficient implementation without multiple conversions of data

• Kernel API for property access?
• SIMD / Java Vector API ?
• Early filtering and stop when top-k is reached or threshold not passed
• Stream results out

Compare large scale performance, with say 1000 / 10k nodes compared with the genai.vector.consine function.