fix(runner): prevent subprocess pipe deadlocks

[March-77]

Summary

• drain stdout and stderr concurrently while commands run, while retaining at most 1 MiB per stream
• run commands in isolated process groups and apply bounded cleanup on success, timeout, and caller cancellation
• cover large exec/shell output, timeout output, stdin, cancellation, and descendant-process cleanup

Validation

• targeted pytest on Windows compatibility harness: 9 passed, 3 POSIX-only tests skipped
• pycln --check on both changed Python files
• isort --check-only and yapf --diff on the new test module
• python -m compileall on both changed Python files

The POSIX shell/process-group regressions are included for Linux CI; the local environment is Windows and had no usable Linux container runtime.

Fixes #53
Refs #62

[March-77]

Revalidated on 2026-08-23 against main 469de08:

• The PR remains clean and mergeable; no reviews or CI checks are currently present.
• Main still waits for piped subprocesses before draining stdout and stderr, so the reported deadlock path remains unfixed there; the runner interface is unchanged.
• All 12 targeted Python 3.11 subprocess tests passed, covering large stdout and stderr, bounded capture, timeout output, stdin, cancellation, and descendant cleanup.
• PRs feat(pytest): support extra command-line arguments #81 and fix(runner): prevent subprocess pipe deadlocks #82 are independent: they share the same base and modify disjoint files.

No code-only refresh commit was added. Maintainer review would be appreciated.