In a future with a more complex LN protocol wrt fees (currently the opener pays everything, which is a bad incentive for such entities to directly open a channel), the exchange could just open the channel and pay the opener through the first commitment transaction pair ?
Sure, as long as the exchange doesn't have any issues (regulatory etc.) to connect to LN.
By signing a transactiono using ACP|NONE you are giving the control not only to your customer but to anyone who claims it, so basically the first miner to include the tx

.
Ah, uh. I may have understood something fundamentally wrong then. So SIGHASH seems to refer always to the signing of complete transactions, not just utxos? But this contradicts what I understood in Sood's blogpost ...
I quote the part of the
blogpost abouth Sighash I misunderstood perhaps:
SIGHASH_NONE - This one is a bit more confusing. On the face of it, it seems like youre burning money by not signing any outputs. Indeed, if you create a tx with just a single input and sign it with SIGHASH_NONE, the miner would be able to simply change the output to one that they control. This is mostly designed to be used in scenarios where more than one party is contributing inputs. At that point, such a signature essentially means I agree to spend my money, provided all these other people spend theirs too. It is expected that one of the other signers will then use SIGHASH_ALL to secure all the outputs of the transaction, and send the money to a mutually agreed output set.
I interpreted this that way, that you could send a "pre-transaction" with your UTXO as an input, signed with SIGHASH_NONE and SIGHASH_ANYONECANPAY via a secure communication channel to another person and this person could add his input and sign it with SIGHASH_ALL and broadcast it. This person in my example from above would be the customer of the service/exchange. So miners would only get access to the transaction once the customer would have "completed it", and then it's late for them to "grab it" as it was already signed with SIGHASH_ALL.