Copy: Parallelize copy_volumes_with_progress with FuturesUnordered

Cross-volume batch copies of 3+ items now run up to
min(src.max_concurrent_ops(), dst.max_concurrent_ops(), 32) streams in flight.
Below the threshold (or when both sides return 1), it falls back to the
original sequential loop unchanged.

- Shared progress state: files_done / bytes_done / last_progress_time are
  hoisted to Arc<AtomicU{usize,U64}> / Arc<Mutex<Instant>> so tasks update
  them without clobbering each other. The 200 ms throttle keeps events at
  `<= 5/s` regardless of task count.
- Abort on first error (F10): the driver returns Err on the first failing
  task; dropping FuturesUnordered cancels the rest, and each backend's
  writer.abort() / delete-partial path handles temp cleanup.
- Concurrent partial cleanup: each in-flight task's destination path is
  tracked in a shared Arc<Mutex<Vec>>; cancel/error deletes all of them
  (rollback includes them, pure cancel just cleans them).
- Conflict resolution stays pre-dispatch (F14): Stop-mode prompts block
  the whole batch, same as before, so racing tasks can't double-prompt.
- Cancellation check at task start AND between chunks inside copy_single_path
  (unchanged per-file semantics).

Author: vdavid