Thanks Corentin for the paper. I hope this feedback helps the discussion.
With respect to the contents of multicharacter literals, the paper does not give much motivation for disallowing numeric escape sequences which fit within a single unsigned char. Also, the wording says "shall be a member of the basic literal character set": this property of "being" is rather ambiguous in terms of authorial intent regarding the treatment of UCNs, etc. that designate members of the basic literal character set (a name for something is usually not the same as the thing it names).
With respect to the new encodability restriction for strings, I believe that unevaluated strings should not be treated the same way as strings that need to be translated into a literal encoding. I think we may need to advance P2361 ("Unevaluated strings") first.