Ah, so I guess the rule could be: blocks (with more than one transaction) at or above 1,983,702 MUST have a witness commitment. That’s plenty of time to make sure all mining software does that by default.
This is probably the only new soft-fork rule that miners could violate by accident. Not requiring a witness commitment for “empty” (only a coinbase) blocks makes it safer, because who knows what custom code is out there to SPV/spy-mine in the brief interval between when a new block is seen and when it’s fully validated and the new template is ready.
There isn’t, unless I missed it: op return - When was the first OP_RETURN output in a coinbase? - Bitcoin Stack Exchange
(actually that’s not exactly what you asked, but at least there’s no OP_RETURN use in the coinbase before BIP30 and BIP34 kicked in)
This is not necessary. As explained in validation.cpp:
// Block 490,897 was, in fact, mined with a different coinbase than
// block 176,684, but it is important to note that even if it hadn't been or
// is remined on an alternate fork with a duplicate coinbase, we would still
// not run into a BIP30 violation. This is because the coinbase for 176,684
// is spent in block 185,956 in transaction
So a re-org would have to go all the way back to 185,956. That’s buried below multiple checkpoints. And although we might get rid of checkpoints completely, many nodes would consider a reorg that deep a hard fork.