Add Initial LLM Extraction for Predator Diet Survey Metrics

[SeanClay10]

Summary:
Implements a minimal LLM-based pipeline for extracting key metrics from preprocessed predator diet survey texts using a Ollama client.

Changes:

• Added LLM script with structured extraction using Pydantic schemas
• Extracts: species name, study location, study date, empty/non-empty stomach counts, and sample size
• Includes post-processing validation to correct LLM calculation errors and compute fraction of feeding predators

Usage:
python src/llm/local_llm.py data/processed-text/paper.txt

Next Steps

This provides the foundational structure for LLM-based extraction. Responses are not the best right now due to preprocessing difficulties and the sheer number of tokens that are sent to the model from the long papers, so changes to the prompts and how the preprocessed text is passed to the LLM needs to be looked into.

[raymondcen]

Everything looks good. The only thing I can really pick out from this are better prompts but we can always change that down the road if we find a better model. We can also just use AI to generate a better prompt. Besides that it looks great.