ANG-008:Similarity computation, edge creation & SQLite cache

[yugalkaushik]

• Compares every note's vector with every other note's vector, using cosine similarity (dot product on normalized vectors).
• For large note collections, uses Joplin's own search instead of comparing every pair one by one, so it stays fast.
• Turns raw similarity scores into a 0 to 1 scale, and keeps only each note's top 5 matches.
• Gives a small score boost when two notes share a tag.
• Gives a small score boost when two notes are already linked to each other.
• Gives a small score boost when two notes were created close together in time.
• Has a safety floor so tags alone can never create a fake connection only real similarity or a real link can.
• Turns the similarity scores into graph edges, on top of the existing link and tag edges.
• Saves each note's vector in a local SQLite database so notes that haven't changed don't need to be re-fetched next time.
• Cleaned up the code so each function does one clear job as suggested