fix(runner): use select_all for true concurrent fail-fast

[0xReLogic]

Summary

Fixes the follow-up concern raised in #58 the collect_parallel_results function was awaiting task handles sequentially, meaning a slow task at index 0 would block detection of a fast failure at index N.

Changes

• Replace sequential for handle in handles loop with futures_util::future::select_all for true concurrent monitoring
• As soon as any task fails, all remaining siblings are aborted immediately regardless of their position in the list
• Add test: slow task first, fast failing task last validates the concurrent fail-fast scenario
• Add test: empty handle input edge case

Tests

All unit tests pass:

test collect_parallel_results_succeeds_when_all_tasks_ok ... ok
test collect_parallel_results_handles_empty_input ... ok
test collect_parallel_results_aborts_remaining_tasks_on_failure ... ok
test collect_parallel_results_aborts_slow_first_task_when_later_task_fails ... ok

Closes the improvement suggested in #58.