Add model prediction and rule integration

[Kaushik-Kumar-CEG]

Adds reusable APIs for model-assisted ScanCode rule updates.

The APIs load eligible rules, validate model predictions with ScanCode, prepare complete updates without changing the input rule, and atomically write the exact source path. Unsafe candidates, protected URLs and filenames, repeated phrases, overlapping phrases, stale files, and unsafe model artifacts are rejected.

Example:

from scancode_required_phrases.model_rules import load_prediction_rule
from scancode_required_phrases.model_rules import predict_rule_candidates

rule_path, rule = load_prediction_rule("path/to/example.RULE")
result, candidates = predict_rule_candidates(rule, predictor)

Test it:

./configure --dev
python -m pip install -e ".[inference]"
venv/bin/pytest -q tests/test_model_rules.py tests/test_inference.py
venv/bin/ruff check src tests

This PR is stacked on PR #2. PR #4 uses these APIs for the user command.