When I persist an aggregate in my domain, there is a very good chance I also need to publish events to a durable service bus - and I absolutely need those two operations to be transactionally consistent.

Sure, there are workarounds, like persisting the events to the same storage as the aggregate, and then having a worker thread read, publish and remove them, but that introduces a huge amount of complexity and raises all sorts of questions, like what happens after a restart and how this affects federations and sharding.
And seriously, once you finally get all this infrastructure implemented, chances are that your performance ends up being a lot worse than if we just had native support for distributed transactions.

So, while I realise that distributed transactions in a cloud environment is inherently problematic, I do think it is an absolute requirement.
We need transactional consistency between persisting to storage and publishing messages.