The fundamental premise behind the current text/paragraph proposal is that
we can't stop vendors from generating this stuff, so let's at least
attempt to get them to label it so the recipient can fix it without
breaking other things.
Your counter-proposal is based on the premise that the vendors who are
generating this stuff are willing to add code to make it palatable to
Internet users. Given that downconversion to text/plain is simpler than
your proposal, and that there is an existing text media type which encodes
paragraph semantics in a human friendly way (RFC 1896), I suspect your
premise is not correct.
Your proposal is creative and might have been a good idea six years ago,
but I don't think it addresses the underlying problem today.
- Chris