Created an attachment (id=390741)
patch ver. 4 (different approach)
Ok, so I came up with this idea instead. Since what we really want here is just a method to override nsContentBlocker's policy for any specific image, and creating a whole api that would be element-based would be pretty crufty (the override would only last as long as the element does, so it wouldn't be all that useful), I think this "mime" passing would do better.
Basically, I'm abusing aMimeGuess (which isn't used for images) and sending in a "message" instead to the content policy. nsContentBlocker catches this message and returns ACCEPT for that image. This makes for a leaner and cleaner patch, plus it has the added benefit of allowing other content policies to opt in and override their policy. This should also be less perfy (no iteration, just a simple string comparison).
Let me know if this will do. I'm leaving the other patch unobsoleted, as this is my only other option. Creating a whole api to ignore policies per element seems like a huge waste to me, so if this won't be accepted, my only other bet is the previous approach. Or new suggestions, which I'm not expecting to float in anytime soon...