feat: Expose use_structured_output in SimpleKGPipeline constructor

[arthurfantaci]

Problem

LLMEntityRelationExtractor already supports use_structured_output=True (as of v1.13.0), which provides guaranteed JSON schema conformance for entity/relationship extraction. However, this parameter is not exposed through SimpleKGPipeline's constructor.

Users of SimpleKGPipeline must either:

1. Drop down to the low-level Pipeline API to set use_structured_output=True on the extractor
2. Accept prompt-based JSON extraction (which works reliably but lacks schema guarantees)

Proposed Solution

Add use_structured_output: bool = False as a constructor parameter to SimpleKGPipeline, and pass it through to _get_extractor():

# In SimpleKGPipeline.__init__:
self.use_structured_output = use_structured_output

# In SimpleKGPipeline._get_extractor():
return LLMEntityRelationExtractor(
    llm=self.llm,
    prompt_template=...,
    use_structured_output=self.use_structured_output,  # <-- new
)

This is a ~1-line change in _get_extractor() plus the constructor parameter.

Context

We're building a production GraphRAG pipeline using SimpleKGPipeline and would benefit from structured output for more reliable extraction. The current workaround (low-level Pipeline API) adds significant complexity.

Versions

• neo4j-graphrag: 1.13.0+
• Python: 3.13