fatxpool: ready_at_light optimization

[michalkucharczyk]

I still think that the following should be faster:

let mut to_process = Vec::new();
while at.number() > last_finalized.number() {
    if self.view_exists(at.hash()) { break }
    to_process.push(at.hash());
   at = at.parent();
}

No need to get a tree route down to the finalized block.

Originally posted by @bkchr in #4639 (comment)