Integrate sentiment analysis pipeline

[justinmadison]

Summary

Add a sentiment analysis step to classify each normalized article as positive/neutral/negative.

Motivation

• Enables tracking of sentiment trends for entities and topics over time.
• Supports alerting on sudden sentiment shifts (e.g. negative spikes).

Scope

In scope: implementation, tests

Acceptance Criteria

• analyze_sentiment(text) returns a non-empty label and numeric score.
• sentiment_task(article_id) stores a "sentiment" field on the article.
• CLI command runs without errors and prints the sentiment result.
• All tests pass in CI and README clearly documents both execution paths.

Additional Context

1. Add dependencies
   • Add transformers and torch to /nlp/requirements.txt.
2. Core function signature
   • In /nlp/core.py define:

     def analyze_sentiment(text: str) -> dict
     

3. Celery task hook
   • In /nlp/tasks.py register:

     @app.task
     def sentiment_task(article_id: str) -> dict
     

4. CLI entrypoint
   • In /nlp/cli.py expose:

     python -m nlp.cli sentiment --article-id=<id>
     

5. Tests & docs
   • Create /nlp/tests/test_sentiment.py to:
     • Assert analyze_sentiment() returns a dict with label and score.
     • Assert sentiment_task() updates the article document with a "sentiment" field.
   • Update /nlp/README.md with:
     • How to run sentiment_task via Celery
     • CLI usage for the sentiment command