Ok, a couple of things here:
(1) Measuring inside diameter of small bearings with calipers will normally result in a slightly-lower-than-actual result, but 5.9 is too low - if you're getting that something's not right there, even if they (calipers or bearings!) are a bit cheap.
(2) Anything other than 'precision ground' rods are normally a touch oversized (in case you want to machine bits of them down to the nominal). SD probably picked something that had a wide-ish tolerance range that worked for them and saved by buying a lot of it.
(3) Pushing a 6.000mm ID bearing half way down a 6.000mm OD precision ground rod will be nigh impossible at the best of times, so you may not even want a perfect fit here...
(4) Forcing bearings on/off things never helps, but is often fun...
To deal with the rough rod, I'd suggest using a way-oversized bearing (e.g. 10mm ID) for the middle one, and a printed/otherwise plastic sleeve between it and the rod. The plastic sleeve is flexible, so should be pretty easy to push down the rod even if it's a bit out of shape (printing tolerance aside...), and then the bearing can press onto this. Once assembled, it would work a bit like bicycle wheel spokes and be surprisingly stiff, especially for the forces likely involved here. When I get mine I'll have a crack at doing it this way and let you know the results properly, in the interim anyone else is free to try of course.
Much as I don't like imperial things, and love the innate beauty of a good bit of 'well-engineered' as opposed to 'hacked together,' the 6.35mm bearing idea is not so terrible - the bend is always in one direction (towards the y motor belt), so it would just sit tangent in there once tensioned up. Even if it was loose, this would at least guarantee the largest movement of the y pulley rod in the y-axis direction would be +/- 0.175mm at the bearing, which by the time you work out the leverage (bending displacement to be accurate, but CBF integrating that many times) would be a whole bunch less at the y-pulleys (which is what translates directly into backlash in the print) which I suppose is a lot better than without anything - depends how you define 'too much', as always.
SD3. Mk2b + glass, heated enclosure, GT2 belts, direct drive y shaft, linear bearings, bowden-feed E3D v5 w/ 0.9° stepper
Smoothieboard via Octoprint on RPi