Honor --jobs with a bounded parallel executor

Problem

Parallel execution accepts a job count, but the current implementation starts one thread per test. The configured n_jobs value does not bound concurrency, which can exhaust resources in large suites.

Proposed work

• Replace per-test thread creation with a bounded worker pool or equivalent scheduler.
• Respect --jobs N exactly.
• Choose a documented CPU-based default when jobs are omitted.
• Preserve deterministic result reporting and thread-safe state updates.
• Ensure hooks and nested suites have explicit concurrency semantics.

Acceptance criteria

• Active test concurrency never exceeds the configured job count.
• Tests verify limits for one, two, and default worker configurations.
• Reporter events and final totals remain correct under concurrency.
• Hook ordering and thread-safety expectations are documented.