The process was pretty straightforward. I V-blocked the shaft to see if the main length if it was "banana shaped", which it thankfully wasn't. Virtually all of the error was at the collet bell.
So I chucked the spindle up in the althe, in the boss where the two bearings go, just to spin it slowly. An indicator on the lower bearing boss showed about .001" (which was almost certainly runout from the chuck) while the collet bell showed about .005" at the upper (toward the bearing) end, and .008" at the mouth.
To fix it, I chucked it up in the 4-jaw, dialed to 0.000" on the bearing boss, and turned the OD in a couple light passes with an indexible carbide cutter. I had to use a sharp, positive rake insert with a very small nose radius, as I was taking very light cuts (.002" to .003" per pass) and the material is quite hard and tough.
I simply turned it 'til there was no more "interrupted" cut, which wound up being about .009".
Then I just faced the nose, recut the OD chamfer, and very lightly shaved the top of the collet bell to make sure it was square.
As I'd already set the compound over to the angle of the collet- before, using a collet and an indicator to get the angle exactly on- I ran a boring bar up the collet taper to clean up that too.
Once I have the spindle reassembled, I'll use a toolpost grinder to give a light cleanup pass to the taper again, as the spindle rotates in its own bearings- which will assure it'll spin with basically zero runout.