Remove replay_data_copy job and simplify sync

- Remove redundant replay_data_copy job entirely
- Refactor sync job to handle HAF cache fetching directly
- Use copy_datadir.sh from HAF submodule (like balance_tracker)
- Remove broken cache-manager copy logic
- Update dependent jobs (cleanup_pipeline_cache, e2e_benchmark)

This follows the same pattern as balance_tracker where the sync job
handles everything: fetch HAF cache from NFS if needed, copy data
using copy_datadir.sh, run sync, save results to NFS cache.