Ollama integration test (env-gated) exercising the LLM + embedding path

[ronaldmannak]

Motivation

All current tests run offline (HashEmbedder + pattern extractor). The Ollama-backed path (OllamaClient, OllamaEmbedder, LLM entity extraction, LLM answer synthesis, and — once merged — LLM community summaries) has no automated coverage.

Scope

• Add an integration test that, only when an env var is set (e.g. GRAPHRAG_OLLAMA_TESTS=1 + OLLAMA_HOST), builds a small graph with the Ollama backends and asserts ask / lightRAG().ask return non-empty, grounded answers.
• The test must skip cleanly (no failure) when the env var is unset, so default CI stays fully offline.
• Optional: a separate, manually-triggered CI job (workflow_dispatch) that runs the Ollama tests against a service container.

Files (isolated)

Tests/GraphRAGTests/ (new OllamaIntegrationTests.swift), optionally .github/workflows/. No library source changes.

Acceptance criteria

• Default swift test (no env var) skips the integration test and stays green with no network access.
• With the env var + a reachable Ollama, the test exercises the real LLM/embedding path.
• CI green.

Context: follow-up to PRs #2 / #3.