iter-142: session hygiene — daemon stop pid honesty, disk GC, eval ASI fix

[ractive]

Summary

• Theme A (daemon stop false-negative): daemon stop was reporting the proxy daemon's PID as if it were Firefox's, and escalating the port-free wait against that same wrong PID — so a stop that actually left Firefox running could still report success, and a stop that genuinely needed more time to free the port never escalated against the right process. Root cause: launch-record.json was a single global file clobbered by concurrent ports. Fixed by scoping it per port (launch-record.<port>.json) and independently resolving/stopping the real Firefox PID via port-owner lookup.
• Theme B (disk growth): dead-owner-pid temp profiles are now reclaimed by the very next launch, not gated behind the 7-day age threshold (this is exactly what let 62 profiles / 2.7 GB accumulate in a day per dogfooding session 63). Added GC for stale daemon.<port>.throttle.json state files and the legacy port-less daemon.spawn.lock name, and moved the sweep to run on every launch (not just the rare daemon-autostart path) so a session that reuses an already-running daemon still gets swept.
• Theme E (eval wrapper edges): the async-IIFE wrapper's statement-boundary heuristic was ;-only, so ASI-separated (newline-only) multi-statement scripts leaked the wrapper into a confusing syntax error. Now tracks newline-based ASI boundaries too (with string/template-literal and bracket-depth awareness). Also fixed the await path silently discarding a trailing bare expression as undefined — it now auto-returns the last top-level expression, matching non-await completion-value semantics, while leaving earlier statements (and any explicit return) untouched.
• Theme F (partial): added a plain wait --sleep-ms form (with a --time legacy alias) so dogfooders no longer have to shell out to sleep.
• Themes C (auto-consent honesty / BBC CMP coverage) and D (full-page screenshot header dedup), plus the console-locale item from Theme F, are deferred to kb/iterations/iteration-144-session-hygiene-followup.md — each needs either a JSON-contract design decision (Theme C) or dedicated pixel-level before/after verification against iteration-135's stitching fixtures (Theme D), or a non-English-locale Firefox to reproduce against (Theme F), none of which were safe to rush into this iteration. See that plan's "Why these three were deferred" section.

Test plan

• cargo fmt --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test --workspace -q — all clean

• FF_RDP_LIVE_TESTS=1 cargo test-live filtered to live_142_* — all 4 live tests pass against real headless Firefox on the default daemon path:

  • live_142_daemon_stop_pid_honesty::live_142_daemon_stop_no_false_error
  • live_142_disk_growth::live_142_profile_growth_bounded
  • live_142_disk_growth::live_142_throttle_json_gc
  • live_142_eval_asi_await::live_142_eval_asi_await_script

• unit_legacy_spawn_lock_collected (unit) and e2e_wait_sleep_form (e2e) pass

• cargo run -p xtask -- check-iteration-ready --plan kb/iterations/iteration-142-session-hygiene.md --base origin/main — 10/10 PASS (with FF_RDP_LIVE_TESTS=1 for the dogfood gate)## Claims vs code
  <generated 2026-08-11T13:22:33Z by ralph-loop>

• immediate → ✅ matched in diff

• eval → ✅ matched in diff## Claims vs code
  <generated 2026-08-11T13:35:46Z by ralph-loop>

• immediate → ✅ matched in diff

• eval → ✅ matched in diff