There's been a lot of past discussion around this issue:
I'm probably missing a few more.
What's different now? Mostly, that we actually (almost) have the machinery in place to build a relatively simple solution. We didn't before.
This proposal is also significantly more narrowly targeted than previous attempts have been. It's not a big sweeping language change. If you squint, you can almost pretend that we're just fixing the behaviors of a few APIs where they currently have bogus behavior. Obviously, it's actually a bigger change than that, but from the perspective of most user's experience of it, that's about the scope of it.
As a personal note, I'll be on vacation for the next two weeks. Some other folks from the standard library team will be helping with this pitch, but if there are specific IEEE 754 questions that I need to respond to, the latency will be longer than usual. Nonetheless, I'd like to start discussion of this so that we can make the change for Swift 5 if we decide to pursue it.