Add human review workflow for model predictions

[Kaushik-Kumar-CEG]

Adds the complete add-model-required-phrases workflow

It supports one rule, a directory or eligible installed ScanCode rules. The default mode reviews predictions interactively. It also supports read-only prediction, explicit score-based batch processing, dry runs, JSON output, and resumable sessions. The pinned public model works by default without a token

Usage:

python -m pip install ".[inference]"
add-model-required-phrases --rule path/to/example.RULE
add-model-required-phrases --rules-dir path/to/rules --predict-only
add-model-required-phrases --all --batch --auto-score 0.90 --review-score 0.70 --dry-run
add-model-required-phrases --resume path/to/session.jsonl

Test the complete package stack with:

./configure --dev
python -m pip install -e ".[inference]"
venv/bin/pytest -q
venv/bin/ruff check
make doc8
make docs-check

This PR is stacked on PR #3. Checking out this branch provides PRs #1 through #4 together, so the complete package can be installed and tested without manually combining branches