When looking at changes made to a file, the ones most likely to be interesting are those affecting the portion of the file in view.  The second attached patch causes `vc-diff' to place point at the place in the diff that corresponds to point in the file being compared (only if the current source is the endpoint of the diff, of course).

The first patch (starting from a recent master) is preparatory and provides (and uses) a defconst for the main work.

To resolve:
1. Would this need a NEWS entry?
2. Does it need to be customizable?  (M-< is all it takes to "recover the old behavior"...)
3. Is it worth supporting the (very rare) "normal" diff format?
4. Should M-x diff (or any other diff commands) behave similarly?
5. Do context/unified headers lack the (correct) length of the hunk often enough to bother with "donttrustheader"?  (This is the two FIXMEs in the patch.)

Davis