I think generally you should be able to break some feature up into a series of very small tasks that can be estimated. So you may not be able to estimate an end to end feature accurately but you should be able to give rough ETAs for when different functionality will start to trickle in

You don’t do estimation without prioritization. As a consultant, my default way of working is kanban, often setup on Trello. Once I deconstructed the project into features, the first thing I do with the client is arrange them by priority and arrange them into sets for releases.

If sales and marketing need to do planning, then they need to tell what they want to push back to the backlog so that the feature they want are on time.

We do that already, yeah. But they'd like estimates on a lot of issues so they can prioritize better. I believe it's a waste of time though. There is no good solution for this I believe.