The TUI was listing raw JSONL filenames as separate sessions, so when OpenClaw wrote both a bare session file and a topic-qualified file for the same canonical session id, the list showed duplicate rows even though both mapped to one logical LCM conversation. Collapse discovered session files by the JSONL session header id and prefer the topic-qualified transcript when both variants exist. Add a regression test for that duplicate-file case.

Regeneration-Prompt: |
  Fix the lossless-claw TUI so it does not show duplicate session rows when the sessions directory contains both <session-id>.jsonl and <session-id>-topic-<n>.jsonl for the same logical session. The canonical identity should come from the JSONL header's session id, not just the filename stem. Keep existing DB lookup behavior for topic sessions, but collapse duplicate on-disk files into one visible row and prefer the topic-qualified transcript when choosing which file to represent that session. Add a focused test that creates both files with the same header id and verifies the topic-qualified transcript wins.