0.2.21

Fixed

• alembic upgrade head failed in every PostgreSQL project without
  --mcp-client — since 0.2.19, migration 0027 named
  0026_create_mcp_connections as its parent, and that migration is generated
  only with the MCP client, so Alembic stopped with
  KeyError: '0026_create_mcp_connections'. 0027 now follows 0025 when 0026 is
  absent. Nobody noticed because tests/test_migrations.py took any alembic current failure for "no database" and skipped itself; it now checks the
  connection directly, so a broken chain fails it. A new matrix check,
  test_alembic_chain_is_unbroken, verifies every generated configuration has
  one head and no missing parent
• Migrations could not start in projects without an AI framework —
  alembic/env.py imported the conversation models, which are generated only
  with one
• The test suite of a project without an AI framework failed at collection
  — tests/test_services_conversation.py stayed behind after its module was
  removed
• Resetting all non-admin users left their agent memory behind — with
  --memory, UserService.delete_non_admins (used by the seed command's reset)
  bulk-deleted the rows but not the notebooks, which have no foreign key to
  users; single deletion already cleared them. It now forgets each deleted
  user's memory, and a store failure rolls the deletion back
• Deep research could drop an answer given alongside a planning step — text
  sent with a planning or delegation tool call is held back as narration; if a
  run nevertheless ended on that step, the text never streamed although it was
  the final result. It is now sent before final_result. Current Pydantic AI
  continues a run after such a step, so this guards a path rather than fixing
  one seen in practice (#176)
• A conversation resumed on a new connection lost its context — the chat
  session held the conversation in memory only while the WebSocket lasted, so
  after a reconnect the model started from nothing, and switching to another
  conversation in the same session carried the previous one's context over.
  The Pydantic AI and PydanticDeep sessions now load the conversation's most
  recent stored messages (up to 100) whenever they switch to a conversation
  they do not hold