I've got to say that I'm rather uncomfortable with the type-based meaning ascription of this proposal.
That being said, we do need a lightweight way to add such modifiers to arrows (and anything which can appear to the left of an arrows such as fields of data constructors, whether in record notation or not (though I'm fine with requiring GADT notation when not a record, that's what we did with Linear Types)). There is a staggering amount of possible such modifiers which have been conceptualised.
By far, this is the least awful such proposal that I've seen. It's the first time that such a proposal looks even remotely usable.
So the pragmatic in me is inclined to accept as well.
The need for such a lightweight mechanism to annotate terms, though, is quite a bit less clear to me. Because type disambiguation is rather novel in GHC, I'd rather be prudent in this area, and stage the term bits as future work. Keep the design, and revisit it once we have experience with the mechanism for arrow modifiers.