Firstly, being able to extract portions of a text document is significant useful new functionality which alone would justify a new version.
Secondly, invalidating all existing documents is clearly most unwise and overloading/extending xpointer could also be problematic, so a new attribute seems the cleanest solution. It does have the minor disadvantage that when some new (non-XML, non-plaintext) media type becomes prominent, yet another attribute needs to be added. But that probably does not happen very often and the proposed 'textpointer' is sufficiently clear that a new 'foopointer' could be cleanly added in a later version.
Third, ID fixup would be valuable provided that it was under the author's full control. In particular it should be possible to prevent it happening (e.g. because scripts are looking for particular IDs which must be preserved as-is).
Fourthly, with the pre-post communication improvement, is there any impact on xproc?
Thanks for a clear and well-motivated requirements document.
 Too Stupid; Won't Read :) because I hate TL;DR
Chris Lilley Technical Director, Interaction Domain
W3C Graphics Activity Lead, Fonts Activity Lead
Co-Chair, W3C Hypertext CG
Member, CSS, WebFonts, SVG Working Groups