CUDA configury updates

Implement core OOO completion simulation logic (~180 lines):

ooo_init_tracker():
  - Allocate tracking arrays for CQE and completion state
  - Initialize counters and window management

ooo_should_post_retransmit():
  - Return true every 4 WQEs to trigger retransmit posting

ooo_process_completion():
  - IN-ORDER mode: WQEs 1-4 wait for retransmit (linux-rdma#5) to complete
  - OUT-OF-ORDER mode: Each WQE completes immediately on CQE
  - Returns number of newly completed WQEs

ooo_print_statistics():
  - Display completion statistics including retransmit count

This enables comparing head-of-line blocking (IN-ORDER) vs
independent completion (OUT-OF-ORDER) behavior.