Checkpoint: Concurrency

Checkpoint — concurrent job processor (hardened)

Same contract as the project: run_jobs(jobs, workers=3, retries=2) returns {'succeeded': {index: value}, 'failed': [sorted indices]}, retrying failures up to retries extra attempts.

This checkpoint's extra scrutiny: every submitted future's outcome must be observed. A resubmitted retry that nobody awaits loses its result and its exception — a real production bug. The wrong-solution probe fires exactly this mistake.