v3.2.1

Fixed

• The transcript sweeper no longer errors when two sweeps run concurrently on
  the same session. _write_sweep_state now writes to a unique per-process
  temp file, so running transcript-sweep run back to back no longer races on
  a shared temp path and raises a spurious FileNotFoundError (surfaced as
  files_failed). No data was lost — content is staged before the state write —
  but the errors were alarming on a first run. (#165)