This PR was only opened for code archival purposes and potential future reference. See #7144 (review) for more context.

I did spike out an approach for this that uses a Map to store individual header heights, but honestly, at the end of it all when I stood there looking at the final code, I couldn't justify adding the extra complexity for something that I can almost guarantee no consumer is going to use. I'd rather wait to get an actual feature request for it first.

I'll probably still open a draft PR with the working spike and then close it to preserve the code in case some absolute madman out there does want it in the future, but I don't intend on initially shipping with support for it.