Feature: Semantic clustering of Performance Schema query patterns

[peter7775]

Problem Statement

The application already collects MySQL Performance Schema metrics and does query pattern analysis / bottleneck detection. Structurally different but semantically equivalent queries (e.g. same query with different literals/aliases) can currently be counted as separate hotspots.

Proposed Solution

• Normalize SQL statements and generate embeddings (via the EmbeddingProvider port) for statements collected by the existing performance analysis layer.
• Cluster semantically similar queries (e.g. simple k-NN/cosine similarity) before the bottleneck detection step.
• Surface merged clusters in the existing performance dashboard.

Alternative Solutions

Syntax-only normalization (stripping literals), which likely partially exists already; embeddings help with structurally different but semantically equivalent queries (e.g. reordered JOINs) that syntax normalization misses.

Use Case

More accurate, less duplicated hotspot/bottleneck reports for users analyzing database performance.

Priority Level

Nice to have

Implementation Ideas

• Builds on the existing "Predictive performance insights" roadmap item (trend/anomaly engine).
• Evaluate whether embeddings also improve the "Query pattern analysis" feature mentioned in the README.

Additional Context

Depends on: "EmbeddingProvider port". Optionally related to: "Native vector index foundation" if cluster embeddings should be persisted in Neo4j.