Require page rendering for page-level image embeddings

[edknv]

Description

Page-level image and text_image embedding could silently return PDF page rows without image payloads or embedding vectors when extract_page_as_image was not explicitly enabled.

This change treats page rendering as a required extraction step whenever embed_granularity="page" is combined with embed_modality="image" or "text_image". It applies the behavior consistently across direct PDF graphs, automatic dispatch, and mixed-input manifest branches. Page-level text-only embedding remains unchanged.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[greptile-apps]

Greptile Summary

The PR automatically enables PDF page rendering when page-level image or text-image embeddings require raster input.

• Adds the rendering requirement to direct and automatic graph construction.
• Applies the same behavior to PDF branches in mixed-input ingestion.
• Adds graph, manifest, and end-to-end coverage and updates embedding documentation.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
nemo_retriever/src/nemo_retriever/graph/ingestor_runtime.py	Detects page-level visual embedding modes and enables page-image extraction before constructing direct or auto-dispatch graphs.
nemo_retriever/src/nemo_retriever/ingestor/branch_extraction.py	Applies the page-rendering requirement to PDF extraction branches before their outputs enter the shared embedding graph.
nemo_retriever/tests/test_ingest_manifest.py	Verifies that mixed PDF and text ingestion enables page rasterization for both visual embedding modalities.
nemo_retriever/tests/test_pdf_render_params_wiring.py	Exercises the SDK path and verifies rendered page payloads and embedding vectors are materialized.
nemo_retriever/tests/test_pipeline_graph.py	Covers direct PDF, auto-dispatch, visual modality, and text-only graph configuration behavior.
docs/docs/extraction/embedding.md	Documents the combined granularity and modality settings required for page-as-image embedding.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    A[Page-level embedding requested] --> B{Modality}
    B -->|text| C[Preserve extraction settings]
    B -->|image or text_image| D[Enable PDF page rendering]
    D --> E[Extract page_image]
    E --> F[Shared embedding stage]
    F --> G[Rows with image embeddings]

Loading

_{Reviews (2): Last reviewed commit: "Merge branch 'main' into nvbug_6628056" | Re-trigger Greptile}