fix(group): persist group chats and route subchat whispers

[uhhgoat]

Summary

Reopens and rebases the closed #4137 group-chat persistence fix against the new odysseus-dev/odysseus dev branch. The change persists group chat state server-side, hides participant child sessions from normal top-level session views, exposes participants through the parent session disclosure, routes direct child sends as private group whispers, cascades parent moves/deletes to participant children, and preserves legacy/null-owner group state rows through the current owner-filtering behavior.

Linked Issue

Fixes #4038.

Type of Change

• Bug fix (non-breaking - fixes a confirmed issue)
• New feature (non-breaking - adds new behaviour)
• Breaking change (changes or removes existing behaviour)
• Refactor / cleanup (behaviour unchanged)
• Documentation only
• CI / tooling / configuration

How to Test

Manual review flow: create a group chat with multiple personas, reload or switch away and back, confirm the parent restores group mode and participants, expand the parent row to inspect participant children, open a child raw transcript, send a whisper, and confirm the parent mirrors the exchange while child lifecycle actions remain blocked.

Validation run while reopening:

git diff --check origin/dev...HEAD
python -m py_compile routes/chat_routes.py routes/session_routes.py src/session_actions.py src/session_search.py tests/test_group_chat_state_routes.py tests/test_group_chat_storage.py tests/test_session_actions_cleanup.py tests/test_session_search.py
node --check static/app.js
node --check static/js/group.js
node --check static/js/sessions.js
node --check static/js/chat.js
node --check static/js/modelPicker.js
node --check static/js/chatRenderer.js

Local pytest was not available in the bundled Python runtime, so the focused pytest suite is left to GitHub CI.

Checklist

• I searched existing issues and PRs for duplicates before opening the linked issue.
• The PR links the issue it fixes.
• I added or updated focused tests for the changed behavior.
• I ran the validation commands listed above.
• The change is scoped to group chat persistence, participant visibility, and direct child whisper behavior.
• I did not add new runtime dependencies.

[RaresKeY]

I reviewed the complete latest head against current dev and found four issues, including one regression that prevents the main streaming chat route from running.

Findings

issue (chat runtime): Restore the independent incognito request flag

• Problem: The stream handler previously assigned incognito beside the other request flags. This change replaces that assignment with group_internal, but the handler still passes incognito into context construction and reads it throughout the response path. There is no remaining assignment in the function.

• Impact: Every /api/chat_stream request that reaches context construction raises UnboundLocalError before model execution, so ordinary chat UI requests fail globally.

• Ask: Keep the existing incognito parse and assign group_internal separately. Please add a route-level streaming regression covering ordinary, incognito, and group-internal requests.

• Location: routes/chat_routes.py:843, routes/chat_routes.py:1093

issue (data integrity): Make group cascade deletion atomic and check every deletion

• Problem: Group deletion commits each child separately, then deletes the parent, then removes group state. A child deletion returning False is ignored. The underlying delete helper converts database errors into False, so a failed child can be left behind while the parent and group state disappear; a later failure can instead leave only some histories permanently deleted.

• Impact: An ordinary group deletion during a persistence failure can silently orphan participant histories or irreversibly erase only part of the group.

• Ask: Delete parent, participants, and group state in one checked transaction, roll back on any failure, and update in-memory state only after commit. Please add injected-failure tests for a child and the parent.

• Location: routes/session_routes.py:378, core/session_manager.py:542

issue (API consistency): Mirror child whispers on /api/chat as well as /api/chat_stream

• Problem: The child-to-parent lookup and both mirror calls exist only in the streaming route. The supported non-streaming /api/chat route still accepts and persists messages to the same child sessions but never mirrors the user or assistant message to the group parent.

• Impact: API and non-stream clients can successfully send a whisper while the parent group transcript silently loses that exchange, so persisted history depends on the transport used.

• Ask: Apply the same owner-scoped mirroring to /api/chat, or explicitly reject group-child use there if it is unsupported. Please add an API-level regression for both directions and for avoiding duplication on internal group requests.

• Location: routes/chat_routes.py:701, routes/chat_routes.py:1261, routes/chat_routes.py:1756

issue (accessibility): Put the whisper destination in the accessible name

• Problem: The button's visible text and accessible name remain Whisper; only its title changes to Whisper to <participant>. A title is not a reliable destination announcement when the button already has child text.

• Impact: Screen-reader users cannot reliably determine which participant will receive the next whisper before activating the control.

• Ask: Update an aria-label or associated accessible description with the active participant and cover the state transition in a DOM/browser accessibility test.

• Location: static/index.html:1174, static/app.js:970

Validation

• 35 focused persistence, storage, lifecycle, and search tests passed in the secretless runner.
• Changed Python compilation, all changed JavaScript syntax checks, and git diff --check passed.
• A fresh conflict check against current dev at e7eddbae passed, and source inspection confirmed that all four findings remain present.
• All visible current-head checks pass, but none exercises the broken streaming runtime, failed child deletion, non-streaming whisper path, or accessibility tree.
• Browser/mobile group interaction and restart migration behavior remain to be checked after the fixes.

[uhhgoat]

Addressed all four findings in c2480d11 (and merged current dev at 17ee856d):

• restored independent incognito and group_internal parsing, with route-level ordinary/incognito/group-internal coverage;
• staged parent, participants, messages/documents, gallery state, and group state in one checked transaction; rollback tests now cover a child returning False and an injected parent failure, while memory eviction and file unlinking occur only after commit;
• mirrored both user and assistant whispers through /api/chat, with group_internal suppressing duplicates;
• added a participant-specific aria-label plus DOM state-transition coverage and a local browser smoke check.

Validation: 5053 passed, 4 skipped for the full pytest suite; focused/adjacent regressions were 47 passed; node --check and git diff --check passed. GitHub CI is running on the pushed commit.

[github-actions]

⚠️ PR description is complete; validation evidence is still outstanding

Changed-file classification: UI-sensitive.

Author-reported runtime / visual state

• App/runtime validation is not author-attested. Check the run box only after running it, or check the explicit not-run box and describe the gap.
• The screenshot/clip checkbox is not checked for this UI-sensitive change.
• The Screenshots / clips section does not contain an actual attachment or link.

Checkboxes are author attestations. GitHub Actions results remain the execution evidence for CI; this check does not prove that a local command ran.

---

This comment updates automatically when the description or changed files change.

[uhhgoat]

Fixed the unstable pytest check in commit 060114a and merged current dev.

The group-chat regressions had drifted from two current route interfaces:

• the shared foreground-routing query double did not implement all(), which the owner-scoped group-child lookup now uses;
• the non-streaming whisper test still mocked llm_call_async even though /api/chat now uses llm_call_async_with_route_fallback and returns requested/actual route provenance.

The focused updates keep production behavior unchanged while exercising the current fallback-aware path.

Validation:

• focused Docker routing/group-chat suite: 107 passed;
• Python compileall, all 7 PR-specific JavaScript syntax checks, and git diff --check: passed;
• GitHub: 18 checks passed, 1 expected SARIF-upload job skipped, 0 failures;
• PR is mergeable and clean.