1. Is there a particular XSL list you refer to?
Oh, indeed. See http://www.mulberrytech.com/xsl/xsl-list/. *The* XSL list, IMHO.
2. ... we process hundreds of thousands of SVG images. ... those attributes do not prevent rendering, but they do prevent validation.
Oh. Wow. Well, I guess it is quite important, then. But I thought you said you were using RNG? I'm guessing the final validation is against a DTD, or else they wouldn't be a problem. (Why? Because in the XDM those aren't attributes, they're namespace declarations. DTDs don't know about modern XML, so they look like attributes to a DTD.)
5. I'm not an XSLT programmer, so this is not surprising, even though it is unfortunate. All I can do is try something, see if it has a positive result (makes my output more like I want it), and go from there. I'll study your changes, and try to incorporate them.
Oh dear; please realize no insults intended. Trial-and-error is a tried and true method, one on which I certainly rely heavily. I was just in a bit of a rush, trying to get mail off to you before I had to go play soccer dad, and up to my eyeballs in projects, here. (One of which is getting ready for an upcoming workshop on XSLT for digital humanists, which is why XSLT programming style is on my mind :-)