Re: delta encoding and state management

Note that in general compression won't do better than minimal
encodings of header names and values.  I say "in general" because, as
we saw in the minimal date/time discussion, we can sometimes further
compress relative to some state (e.g., normalizing time to a newer
epoch), but even so, generic compression algorithms won't figure out
such things.

Therefore it seems best to focus on minimal encodings.