compress_session.py Phase 4 builds to_del from both processed files
and every skipped file. Phase 1 marks files younger than 2h as
'recent(<2h)' (line 175) so their still-active writer is not
interrupted, but Phase 4 then deletes them anyway.
Filter the skipped-loop by reason so 'recent' files are preserved.
Processed files and other skip reasons (dup, compression error)
still proceed through deletion as before.