workflow: fail runs on hook conflicts

[fantix]

This PR fixes a bug that, await UserHook.wait(token=SAME_TOKEN) may hang forever under a race condition.

When two active runs claimed the same token, the World correctly persisted a hook_conflict event for the loser, but the Python handler ignored the event returned from hook_created and acknowledged the queue delivery. Nothing then woke the run to pass that conflict back to the workflow.

The handler now performs another replay pass inside the same delivery, after receiving the hook_conflict from the world, to fail the run. BaseHook.wait() may now raise a newly-added vercel.workflow.HookConflictError (new public API), carrying the conflicting token and an optional conflicting_run_id

---

Summary

• Detect hook conflict events returned from hook creation and cold-replay the workflow immediately.
• Add a public HookConflictError with Python/TypeScript-compatible serialization.
• Cover conflict liveness, pending waits, typed causes, and non-conflicting registration behavior.

[msullivan]

Looks good!