Stats cleanup

[qinsoon]

• We should add comments (merely copying from the old MMTk would be fine)
• LongCounter should be private. Timer is public.
• Clarify 'phase' - it is not the scheduler phase. It is about gc/mutator. However, we do not want to make assumption that those two are mutually exclusive (concurrent GC).
• Consolidate WorkerLocalStats with stats.
• Possibly remove lock for each Counter.

[qinsoon]

• For Timer (LongCounter<MonotoneNanoTime>), print_total() and get_total() return value in different units (ms vs ns).