Misc performance optimizations and bug fixes

[santhnm2]

List of changes:

• Comment out the preconditions code for now - only necessary when debugging for real cluster.
• Cache values in JobIdPair instead of computing them on every function call.
• Keep a running total_steps_run data structure instead of computing it on each function call.
• Remove unnecessary add_to_queue call in _done_callback.
• Delay recomputing allocation until necessary (when we determine the jobs to schedule in the next round).