Do not use the benchmark index when computing benchmark request queue

[Kobzol]

This removes one half of the usage of the benchmark index. Now when loading the status page or computing the queue for deciding which requests to enqueue, we only load pending requests from the database, together with the information whether their parent is completed or not, and we no longer have to load all completed requests from the DB.

This leaves only one usage of the benchmark index, when determining whether to add master/release requests into the database.

[Jamesbarford]

This leaves only one usage of the benchmark index, when determining whether to add master/release requests into the database

Would one of the database conditions we've added catch this? Also if we need a Vec<_> to check something prior to inserting something into the database then we should be able to do that in a query? Perhaps not in this PR but it theoretically seems reasonable unless I've missed something.

With one exception; when starting the new system we need to temporarily have a manual check for the parent_sha and make it Null if it does not exist as it's own request.

[Jamesbarford]

LGTM