Fixes #57

[Iiridayn]

cursor.each is called lazily, which means we have a race condition between destroy and _onChange. To resolve that, test if destroy has already run, then give up early processing the change list.

[coveralls]

Coverage increased (+2.1%) to 90.05% when pulling 530fe33 on Iiridayn:fix-#57 into 3f15ade on deepstreamIO:master.

[Iiridayn]

All tests pass on my local machine - I suspect some degree of non-determinism in the tests therefore.