Simply order by offset also when not doing brute force to fix #1208

[BobLd]

The regression was "introduced" in 5a6b3970 but removing the code now makes the Issue874 test fail. As a result it appears this commit just surfaced another issue (cc @ricflams)

@EliotJones do you remember why you introduced custom order logic when not doing brute force? I'm now using the same logic for both paths and it looks like it works.This was originally part of commit 0afe021a

[EliotJones]

It has been a while since I wrote the code so I don't remember the context but in general the Xref streams and tables form a linked list (that you can only navigate in reverse by /Prev) so the custom ordering logic respects the order of that linked list rather than the file order.

This might be the case where an incremental update to a file touches an existing object and the update logic writes the new Xref before (in file order) the existing file's Xref. Does this happen in reality? I'm not sure.

Ultimately I don't think we need to respect the linked list in most scenarios so file order is probably a safer default.

[BobLd]

@EliotJones thanks a lot for getting back to me - understood. I'll merge as is and let's see if in breaks any downstream unit tests