Add me as a +1 for the idea that content-type is important for this.
I tend to agree with the arguments given so far. Namely, for some
important use cases you're going to want to know the content type and
guessing is really a bad idea.
That said, there are security considerations associated with specifying
a content type too. I'm particularly imagining a situation where some
sort of deep inspection security appliance uses a different procedure
for deciding what kind of foo it is than the ultimate application. One
guesses based on byte stream another looks at a content type. This is
well known; it's not new; it's probably even so documented that any
reasonably current set of MIME security considerations already includes
a reference.
I agree that your draft does not use the authority portion of a URI
consistent with section 3.2 of RFC 3986. The authority separates the
namespace exactly the way it doesn't in your scheme. It's a naming
hierarchy. My main concern is whether the relative reference algorithm
described in section 5/4.2 of RFC 3986. In particular take a look at the
last part of section 1.2 of RFC 3986 regarding the disallowing of
/. Consider how you want relative references in an HTML document
resolved through a ni: URI to work. I don't think your use of authority
provides good results. However I'm not sure that better results would be
achieved without hierarchy. I hope though that these comments will help
inject some ways of reasoning about authority that are less mystical and
that lead to more practical discussion.