feat: parallel dgoss execution via shared ParallelShellExecutor

[ianpittwood]

Summary

Adds bounded-concurrency parallel execution of dgoss test commands, built on a new tool-agnostic posit_bakery/parallel/ module so other plugins (hadolint, wizcli) and future sequential builds can adopt the same pattern.

• posit_bakery/parallel/ — ShellTask, ShellResult, resolve_max_workers, and ParallelShellExecutor (a ThreadPoolExecutor-backed runner). Worker threads only run their subprocess and return captured bytes; the main thread drains completions, drives an optional Rich live status table, and invokes a per-result callback — so consumers mutate shared state without locks. Results are returned in input order; spawn failures are captured (never crash the run).
• Concurrency control — --jobs/-j CLI flag on bakery dgoss run → BAKERY_MAX_CONCURRENCY env / SETTINGS.max_concurrency → default of 4 (modest, since each task is a Docker container). Malformed env values fall back to the default with a warning instead of crashing the CLI.
• Streaming/logging — the live Progress table shares the same stderr_console as the logger, so per-task detail (flushed on completion, on the main thread) renders cleanly above the live region with no interleaving. The table auto-disables on non-TTY/CI, --quiet, or single-task runs, falling back to plain logging.
• dgoss — DGossSuite.run() now dispatches through the executor while preserving its exact (GossJsonReportCollection, errors) return contract; per-result parse/persist/classify logic is unchanged.

Design spec and implementation plan live under docs/superpowers/ (gitignored, not committed).

Test Plan

• just test — 1639 passed
• New unit tests: executor (ordering, bounded concurrency, spawn-failure capture, env/cwd, main-thread callback, live-display policy), resolve_max_workers, settings env parsing (incl. malformed/empty fallback)
• dgoss suite tests: mocked parallel happy-path, jobs→max_workers wiring, spawn-failure → BakeryDGossError
• CLI test: dgoss run --jobs 2 forwards jobs=2 through to the suite
• Manual smoke (requires Docker + built images): bakery dgoss run --jobs 2 -v shows the live status table + per-target detail, then the Goss summary

🤖 Generated with Claude Code

[github-actions]

Test Results

1 673 tests  +39   1 673 ✅ +39   8m 38s ⏱️ +35s
    1 suites ± 0       0 💤 ± 0 
    1 files   ± 0       0 ❌ ± 0 

Results for commit 816147c. ± Comparison against base commit 6bc441c.

♻️ This comment has been updated with latest results.

[ianpittwood]

Follow-up: per-task timeout + responsive interruption

Added on top of the initial parallel-dgoss work (4 commits, b7b5c1c..bf28a59):

Responsive interruption — ParallelShellExecutor now launches each child with start_new_session=True and tracks live processes in a lock-guarded registry. On KeyboardInterrupt/BaseException it cancels queued tasks and signals the whole process group (SIGTERM → 10s grace → SIGKILL) of every in-flight child, then re-raises. This fixes the prior behavior where Ctrl-C drained the full queue and could leave docker/dgoss containers running. Group-signaling (vs. signaling only the dgoss wrapper) is what prevents orphaned docker grandchildren; SIGTERM-first lets dgoss's EXIT trap remove its container. A registration-race window (a child spawned exactly as shutdown begins) is closed deterministically via a shutdown flag checked under the same lock.

Per-task timeout — ShellTask.timeout enforced via Popen.communicate(timeout=…); on expiry the child group is stopped and the result flagged timed_out. Exposed on dgoss as GossOptions.timeout (default 900s = 15 min), resolved through DGossCommand → ShellTask (0 disables). A timed-out target is reported as a run-failing BakeryDGossError.

Tests — real-subprocess executor tests for timeout, SIGINT cancellation, whole-process-group termination (heartbeat-divergence), and the shutdown-race path; dgoss suite tests for timeout→error and timeout plumbing. Note: this also repaired two dgoss suite tests that the subprocess.run→Popen switch had silently broken (they now mock ParallelShellExecutor.run). just test: 1649 passed.