Good point that committing to only one input does not have the pinning issue, although I suppose the same can be solved with standardness policy. I don’t know if policy rules + replacement rules can solve it in general for an arbitrary number of inputs - it seems like an interesting question.
Malleation in BIP-119 mostly refer the TXID, so it’s out of scope in the discussion above.
The half-spend problem is a good point, though. CCV doesn’t have it because of the cross-input amount logic. For a CTV-like opcode, the only other solution I see is to only allow one input to use the opcode - which might be sensible, but requires some cross-input logic (albeit very simple).