Refactor: unify two-phase dispatch and fix sync_start spin loop

- Refactor AIC/AIV/MIX pending dispatch into a single dispatch_shape()
  that iterates IDLE then PENDING phases for each shape, replacing the
  three separate idle/pending dispatch loops
- Add CoreTracker::DispatchPhase enum and get_dispatchable_cores() as
  a unified query for both phases
- Rename get_idle_cluster_offset_states -> get_idle_core_offset_states
  and consolidate get_pending_only_{aic,aiv} into
  get_pending_core_offset_states, adding MIX pending support (all 3
  cores must be running + slot free)
- Fix: break out of dispatch_shape loop when entire batch was requeued
  (all sync_start); previously caused infinite spin in PENDING phase

Author: zhusy54