Add a Troubleshooting page: stranded-join recovery and the wait-node vs WAITING distinction

Follow-up to #3608046: Fix 1.0.0-alpha7 audit findings: completion/timeout/migration concurrency, join-stall recovery, and re-offer notification. Documents the operator-facing recovery paths that the audit-cycle work added, and clarifies a terminology trap. Docs and one code comment only; no behaviour change; phpcs and cspell clean.

• New Troubleshooting page (docs/troubleshooting.md, under "Running workflows") covering two rare asynchronous states: an instance stranded at a timeout-less join whose awaited branch was cancelled (recover by making the join a timeout join, via an edit or a migration, then arming it), and an instance that briefly shows running with no live token after concurrent branches end (self-heals via the cron reconciliation pass, no action needed).
• Clarifies on WorkflowEngine::isStalled() that the WAITING token status is join synchronisation only: a wait node and every external interaction park as PARKED, not WAITING, so an instance idling on external input is never misread as stranded; and that the stall detection is a heuristic that only logs or arms a declared timeout, never acting destructively.