Add dataset and composite required phrase commands

[Kaushik-Kumar-CEG]

adds the package foundation and two installed commands:

build-required-phrases-dataset creates deterministic BIOES training splits from marked ScanCode rules
add-composite-required-phrases uses existing single-license required phrases to update matching composite rules

both commands use ScanCode rule loading, validation, matching and mutation APIs

Usage:

build-required-phrases-dataset --rules-dir path/to/rules --output-dir dataset-output
add-composite-required-phrases --dry-run --verbose

Test it with:

./configure --dev
venv/bin/pytest -q tests/test_dataset.py tests/test_composite_rules.py
venv/bin/ruff check

This is the first PR in the package stack