Multiple wait events fix and machineId supported

[marcopiraccini]

Summary

Fixes two issues surfaced while running the birthday-card-generator example against the local Platformatic cluster:

1. Duplicate wait_* events crash workflow replay. The SELECT-then-INSERT in plugins/events.ts races under concurrent dispatch retries, producing two wait_created / wait_completed rows for the same correlation_id. The SDK then fails replay with Unconsumed event in event log and the run dead-letters.

2. ICC handler registration rejected. ICC PR icc3#788 renamed podId → machineId in the body it POSTs to /api/v1/apps/:appId/handlers. The workflow service still required podId and 400'd every registration, so workflow_queue_handlers stayed empty and queue messages had nowhere to dispatch.

Changes

• migrations/003.do.sql — partial unique index (run_id, event_type, correlation_id) for wait_created and wait_completed. Intentionally does not cover hook_*: createHook async iterators legitimately produce multiple hook_received events on the same correlation id.
• plugins/events.ts — insertEvent uses INSERT … ON CONFLICT DO NOTHING for the deduped types and falls back to a SELECT when the conflict fires, so callers still receive a row.
• plugins/handlers.ts — /handlers accepts either podId (legacy) or machineId (post-#788), keeping the existing contract working while letting current ICC clients register.

[ivan-tymoshenko]

lgtm