feat(compression): adaptive auto transcript compactor

docs/tui-ux-playbook.md

@@ -189,3 +189,18 @@ illustrations:
 - The pipeline reuses the same stub provider harness as the snapshot tests, so
   new scenes should be added in tandem with `chat_snapshots.rs` to keep visual
   regression coverage and documentation imagery aligned.
+
+## 9. Transcript Compression
+
+- The compactor lives under `[chat]` in `config.toml`. Defaults keep
+  `auto_compress = true`, `trigger_tokens = 6000`, and retain the last eight
+  turns verbatim.
+- Strategy is configurable: `provider` summaries call back through the active
+  model (or `chat.model_override`), while `local` uses a heuristic bullet list
+  for fully offline runs.
+- Users can disable the feature per session with `owlen --no-auto-compress`, or
+  at runtime via `:compress auto on|off`. `:compress now` triggers an immediate
+  compaction even when auto mode is disabled.
+- Each compression pass replaces older turns with a system summary annotated by
+  `message.metadata.compression` (strategy, timestamps, token deltas, and the
+  archived message ids) to support audits and future rehydration tools.