Add actor-scoped conversation MCP

[witbrock]

What changed

• adds actor-scoped conversation_list, conversation_get, and conversation_manage MCP tools
• supports bounded transcript/situation reads and rename, hide, unhide, pin, and unpin actions
• stores user-scoped conversation preferences and overlays them in the existing conversation UI
• preserves owner/shared-session authority boundaries and returns effect receipts with canonical read-back

Why

Agents could inspect and manage conversations only through UI-specific pathways, preventing requests such as renaming or hiding recent substantive conversations from being completed through an authorised tool surface.

Impact

The same user-visible conversation operations are now available through bounded MCP contracts without exposing foreign conversations or debug-heavy payloads. Existing UI behaviour remains available and preference changes synchronise with the server.

Validation

• 73 focused backend tests
• 3 focused frontend tests
• Ruff on new Python files
• affected static-JS lint
• git diff --check
• faithful stdio MCP replay covering list, bounded get, pin, unpin, and final canonical read-back

Jira: JVNAUTOSCI-2637