Prevent reruns from inheriting stale browser harness state

[Cheggin]

Rerun keeps the app session id while replacing the assigned CDP target. The harness REPL was keyed only by app session id, so reruns could reuse a long-lived browser-harness-js process that still held CDP state for the prior target.

Scope the deterministic REPL port by target id as well as session id, and make the stock REPL reconnect if an existing CDP socket cannot attach to the assigned target.

Constraint: Rerun reuses the app session id but creates a replacement browser target

Rejected: Kill all harness REPL processes on rerun | broader lifecycle churn and could affect unrelated active sessions

Confidence: high

Scope-risk: narrow

Directive: Keep Browser Harness JS REPL identity aligned with the assigned target when target lifecycle changes

Tested: yarn test --run tests/unit/hl/browserHarnessEnv.test.ts

Tested: yarn test --run tests/unit/hl

Tested: yarn typecheck

Tested: yarn lint

Tested: yarn test

Not-tested: Live Electron rerun from this worktree

---

Summary by cubic

Prevents reruns from inheriting stale CDP state by scoping the browser-harness REPL to the target and reconnecting when attach fails. Keeps the REPL aligned with the active target across reruns.

• Bug Fixes
  • REPL port is now derived from sessionId:targetId and applyBrowserHarnessEnv sets CDP_REPL_PORT with ctx.targetId.
  • If session.use(targetId) fails, the stock REPL closes the socket and reconnects to the assigned target.
  • Hardened Session socket lifecycle: ignore messages from stale sockets, clear state on close, and reject pending requests on teardown.
  • Added unit tests for port scoping and explicit port override.

^{Written for commit 9506969. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 4 files