Refactoring of priorities

[spirali]

This PR cleans up priorities before landing the new scheduler.

• Priotity is now a newtype and represents a priority combined into u64. All previous tuples like (user, resource, scheduler) priorities were removed.
• UserPriority is a newtype that represents a user priority.
• T-level priority was removed (It was hot fixing some priority when switching from B-level computations. We do not have any practical benchmarks that it is really useful. All actually found problems (connected to non-finishing an "almost completed" job) is actually solved via propagating inverted job id into the priority (= older jobs have higher priority).
• Scheduler priority is now dynamically computed, so it is not stored in Task struct, so the struct is now smaller. Is is now cheap to compute it. Moreover the future scheduler stores priority in the priority queue so it would be completely redundant information.

[Kobzol]

Looks reasonable, just didn't understand some of the magic constants :)