I'm just some random person who creates, and promptly abandons random things.
This mostly helps filters written in scripting languages where there can be long startup times, as this feature will start the filter once, and use the same process for all pages going through that filter.
This will be opt-in by a configuration flag since dealing with multiple pages per execution will require specialized filter scripts.
To let filters know when a page ends, and a new one starts, a seperator based on the initial page content (hashed in a cheap way) will be used as a way to make sure the page content doesn't accidentally contain the seperator.
While basing the seperator on the initial page content allows other filters to craft arbitrary seperators, this is moreaccident preventionthan consistency, and is easier to implement and keep track of (hashing only needs to happen once per page, instead of once per step)