Going back and finishing up on old emails....
Actually, the spec does address stream-ID wrapping. There was a slight mistake with it, but it specifies the following:
"Stream IDs do not wrap: when a client or server cannot create a new
stream id without exceeding a 31 bit value, it MUST NOT create a new
stream."
I believe this is a small performance compromise (needing to open a new connection after 2,147,483,648 streams having been created in a single direction on a session), while making it very easy for implementations to test this edge case...
Mike