We should just compute a root over the events at some height (using bmt) rather than storing an events root.

This change will cut ~50% off the commit flow at the end of a block (

), won't limit our ability to service event proofs (can still prove any subset of block events using this bmt), and will simplify recovery logic (don't need to worry about committing event before state).

If we want to retain a global event index for all items, we can store the last index value in the commit metadata (with height) and start event block indexing at that value. This would still allow us to offer an API that walks the event history without any meaningful overhead for the node.