Add select_subset.py for stratified benchmark task sampling

Selects representative subsets across suite effect-size buckets
(high-positive, near-zero, negative, mixed), stratified by language,
difficulty, and codebase size band. Includes --seed for reproducibility,
--power-report for DOE validation, and outputs both JSON (for
run_selected_tasks.sh --selection-file) and plain-text task lists.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: sjarmak