Implement modular benchmarking pipeline for patient graph evaluation

[Vicbi]

Implement modular benchmarking pipeline for patient graph evaluation

♻️ Current Situation & Problem

This PR introduces a new modular benchmarking pipeline for evaluating patient trajectory graphs derived from clinical case reports. It addresses the need to validate and compare different representation techniques (LLM-based reconstructions, BERTScore, structural checks, and trajectory embeddings).

⚙️ Release Notes

• Added LLMReconstructor using DSPy-compatible APIs for narrative generation from graph data.
• Integrated BERTScoreEvaluator to compare reconstructed vs. original text.
• Introduced TopologyValidator for validating DAG structure, timestamp order, and connected components.
• Added TrajectoryEmbedder based on Bio_ClinicalBERT with pooling strategy for per-patient vectorization.
• Added CLI scripts (main.py, batch_run.py) to run the pipeline on single or multiple graphs.
• New visualization utilities support t-SNE, heatmap, and cluster plotting.
• HTML input support for case report text.

📚 Documentation

• Each module includes inline docstrings and logging for clarity and debugging.
• Configurable via config.py to easily switch between LLM backends and embedding models.
• Visuals and results are saved in output/ and output/plots.
• Usage scripts added for both single-graph (main.py) and batch mode (batch_run.py).
• README to be updated in a follow-up PR.

📝 Code of Conduct & Contributing Guidelines

By submitting this pull request, you agree to follow our Coding Guidelines:

• I agree to follow the Coding Guidelines.