Well, I apparently lied a bit. Now that I'm back in the office with access to all my stuff, I've spent some time testing this again today.
When I said Resolve doesn't write the Avid 'ACLR' atom, I was wrong -- it IS writing it. I have no idea why I wasn't seeing it over the weekend at home, but I had so many test files I'd generated that I guess I was looking at the wrong thing at some point at 3am.
In any event, it doesn't change the results (in this case...)
As far as Premiere not showing the right thing when I originally said it should, well I was wrong there too. Premiere
does have its own DNxHD decoder as I said (as of like CS4 or CS5 anyway, I don't really recall exactly) -- but in sitting here testing it, I can only conclude that it's broken, at least with Quicktime-wrapped DNxHD. It seems to always treat the data as if it were RGB, and re-scales the levels to 16-235 (when internally it's already at those levels) giving you the same washed-out look you do with Quicktime Player.
However it seems like if you render to DNxHD as MXF instead of Quicktime from Resolve, Premiere gets
that right. So that appears to be the solution here.
*sigh* I will try to bring this up with Adobe to see what's going on inside Premiere, but it's almost impossible to talk to an actual human being who knows anything at that company.