Add performance regresssion predictor inference feature

[Ali-Sayed-Salehi]

Summary

Adds an MVP inference-only Performance Regression Predictor model.

The predictor fetches a Phabricator diff by diff ID, extracts the commit message from diff metadata, cleans the message, converts the patch into the structured representation used during training, runs a Hugging Face PyTorch sequence-classification model, and returns a raw performance-regression risk score.

This does not add a training workflow yet. The model artifact is expected to be provided through the standard bugbug model artifact path as performanceregressionpredictormodel.tar.zst.

Main changes

• Added the performanceregressionpredictor model implementation
• Added an HTTP endpoint for Phabricator diff prediction
• Added a local CLI for testing with a local model directory and patch file
• Added Phabricator diff metadata / commit message support
• Added local Docker documentation for running the HTTP service
• Added tests for preprocessing, commit message cleaning, Phabricator metadata handling, and HTTP prediction behavior

Testing

Ran relevant unit tests locally:

uv run --all-packages --group test pytest -q \
tests/test_performance_regression_predictor.py \
tests/test_phabricator.py \
http_service/tests/test_performance_regression_predictor.py