fix(tui): persist dashboard/TUI conversations on WS disconnect/restart

[WilsonKinyua]

Problem

Dashboard/TUI WebSocket conversations could contain many events yet be absent from state.db across a disconnect/restart. On teardown, _finalize_session is the safety-net that flushes any un-persisted transcript, but it called:

agent._persist_session(snapshot, conversation_history=history)

where snapshot (agent._session_messages) and history (session["history"]) reference the same list once a turn completes. Inside _flush_messages_to_session_db, every message whose identity appears in conversation_history is treated as already-durable and skipped — so the call wrote nothing. When _finalize_session was the sole persist path (e.g. an in-turn flush hit a transient SQLite failure, then the WS disconnected/the gateway restarted), the whole conversation was lost.

Fix

Flush agent._session_messages via _persist_session's intrinsic _DB_PERSISTED_MARKER dedup — the same contract the gateway-shutdown flush already uses (#13121):

• Drop conversation_history=history (it aliased the snapshot and neutered the write).
• Drop the or history fallback (it wrote a markerless copy of already-durable resumed rows → duplicates).

Markers persist the genuinely-unflushed tail while leaving durable rows untouched (including a resumed-but-not-run session's already-in-DB transcript), so no duplicates are appended.

Scope: reliability slice 1 only — no changes to cron, Telegram outbox, reconnect protocol, or monitoring. Stamping resumed history durable at load time is a noted follow-up (shared with the gateway path).

Test plan

New real-DB E2E tests in tests/tui_gateway/test_finalize_session_persist.py (drive the actual AIAgent flush against a real SessionDB — the prior tests only mocked _persist_session, so they asserted the call but not the write):

• test_unflushed_turn_survives_disconnect — completed turn whose flush didn't durably land is persisted on disconnect (was 0 rows before the fix).
• test_resumed_session_not_reflushed_as_duplicates — resumed, no turn run → no duplicate rows.
• test_resumed_then_run_turn_not_duplicated — resumed and runs a turn: in-turn flush marker-stamps the loaded prefix, finalize re-appends nothing.

Mock-based contract tests updated to the corrected call shape.

Ran locally:

• pytest tests/tui_gateway/test_finalize_session_persist.py → 14 passed
• Broader: test_finalize_session_persist + test_13121_shutdown_inflight_transcript_flush + test_7100_transient_failure_transcript + test_42039_duplicate_user_message + test_hermes_state + test_empty_session_hygiene → 386 passed

Infographic

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.