fix(engine): isolate each event dispatch snapshot

[joeykchen]

A callback that synchronously triggers its own event can overwrite the outer dispatch's shared callback arrays, causing duplicate callbacks and a nil-call panic. Concurrent triggers also race on those arrays. Give each dispatch an ordered callback snapshot captured under the mutex, then invoke it after unlocking.

The cleanup pass shares the snapshot operation across all three event arities and removes persistent scratch buffers, their cleanup helper, unused state and redundant constructor initialization. Subscription order and changes during dispatch retain their existing semantics.

Validation:

• Reentrant regressions fail on the old implementation and pass for Event0, Event1 and Event2.
• Callback mutation, panic recovery and concurrent dispatch tests pass; the engine event package passes the race detector.
• Runtime consumer tests, pure-engine checks and root wasm compilation pass; runtime output is regenerated twice for consistency.

[fennoai]

Reviewed the complete fixed diff for the event snapshot ownership change, including reentrant dispatch, subscription mutation, panic recovery, concurrent triggers, and the generated type snapshot. No actionable correctness or regression findings remain. Verified with go test ./pkg/spx/pkg/engine, go test -race ./pkg/spx/pkg/engine, and go test ./pkg/spx/....