That is not a feature. It's really frustrating. Probably the most frustrating day to day bit of AWS ES I know of. 
It appears that if the delete is continuing in the background Curator notices that as a
2018-05-08 14:18:18,029 INFO Snapshot activity detected in Tasks API
and I get into the retry logic, which is what I want.
In all cases, the delete does actually succeed, and fairly quickly after the timeout happens. Unfortunately, I can't seem to get the single delete_snapshots action to notice that and retry itself, hence the idea of having multiple delete_snapshots actions.