Proper event/data handling on reply in multitest

[ethanfrey]

In #323, we added support for handling SubMsg in multi-test. Namely, execution was only committed to state on success (multi-level caches) and if reply_on matched, they reply was called from the contract that sent the SubMsg.

All state is properly updated. However, I think we don't properly handle events or data when reply is called (combining them with the actual execution).

In fact, with some reflection, I think we may be doing the wrapping in the wrong place.

[ethanfrey]

closed with #328 and #323