EmbeddingProvider port + adapters (OpenAI, Ollama) for AI/vector features

[peter7775]

Problem Statement

Planned AI features need to generate embeddings from text (table/column names, SQL queries). The project currently has no dependency on any LLM/embedding API and follows a strict DDD architecture (internal/application/ports/).

Proposed Solution

• Define an EmbeddingProvider port in internal/application/ports/, e.g. Embed(ctx context.Context, texts []string) ([][]float32, error).
• Implement adapters in internal/infrastructure/:
  • OpenAI (text-embedding-3-small) for cloud deployments.
  • Ollama (nomic-embed-text or similar) for on-prem/enterprise customers who can't call external APIs.
• Add provider configuration to the YAML config (config/config.yml and the examples/ samples), following the same pattern as existing database connection config.
• Cache/persist embeddings so they aren't recomputed on every transformation run.

Alternative Solutions

Hardcode a single provider (e.g. OpenAI only) — rejected, since enterprise customers (the project's dual-license model targets them) often require on-prem solutions without external API calls.

Use Case

Shared infrastructure for all downstream AI features (semantic search, rule suggestions, query clustering, text-to-Cypher).

Priority Level

Important

Implementation Ideas

• Follow the existing ports & adapters pattern used for database connections in internal/infrastructure/persistence/.
• API keys via env vars / Viper config, never hardcoded in YAML.

Additional Context

Related to: "Native vector index foundation" and every feature issue below.