Request retrying needs to update file offsets in the tried requests

When a document change notification comes in, we currently cancel all outstanding requests and retry them after we changed the document, this obviously causes a problem: Offsets in those requests are now out of sync with the document which can cause a variety of odd issues (and panics). We need to fix them up before retrying again.