Ah yes, I've heard of that before. But if that happens, you can still make your Merkle tree larger and most likely not have collisions anymore, right? So this is clearly a bug in the original implementation, but no deal stopper. (And in theory this could be fixed by using a different way to assign slots.)
I agree. You can always use a higher number of slots, with little space wasted for the merge-mining proofs.
With Armadillo, you will never be able to put an arbitrary number of merge-mined chains into a fixed amount of coinbase space.
(Note that I'm not saying Armadillo is a bad idea, I just previously had the impression when talking to miners that coinbase space is a concern for them or for some of them.)
Yes. The space could be a concern, but the monetary incentive to merge-mine is higher than the cost of the extra bytes.
I think the major problem for miners to add new merge-mined sidechains is not the space, but the one-time integration cost with a new full node. And the periodic maintenance cost to upgrade it.