feat(integrations): AutoGen adapter + integration CI; fix broken crewai/langgraph adapters (Phase 1, Week 3-4)

[tcconnally]

Roadmap Phase 1, Week 3-4: Framework Adapters. Completes the third adapter (AutoGen) and hardens the existing two.

New: AutoGen adapter (integrations/autogen/)

mimir_autogen.MimirMemory implements the autogen_core.memory.Memory protocol (AG2 / autogen-core v0.4+):

• add → mimir_remember, query → mimir_recall, update_context injects a mimir_context block as a SystemMessage, clear → mimir_prune, close reaps the process
• Maps MemoryContent ↔ Mimir entities; persistent stdio session (spawn-once)
• pyproject.toml, README.md, and 7 passing tests (framework deps stubbed)

Fixes uncovered while wiring CI

No CI ran the Python integration tests before, so these were silently broken on main:

• init handshake id bug (crewai + langgraph): the init response was read with req_id - 1 (= 0) but initialize used id 1, so _read_response busy-waited the full 30s timeout on every real session start. Now reads the correct init_id.
• broken test doubles (crewai + langgraph): FakePopen still used the old communicate() API after the adapters moved to persistent write/readline stdio — every test errored with FakePopen has no attribute 'stdin'. Rewrote both fakes to drive the stdio session.
• wrong CLI flags (langgraph): adapter passed a nonexistent --ollama-url and misused --embedding-model (which expects an ONNX model path, not a name). Now maps ollama_url→--llm-endpoint and embedding_model→--llm-model.

New CI: .github/workflows/integrations.yml

Matrix-tests all three adapters (autogen, crewai, langgraph) on every integrations/** change. langgraph installs the real package because it imports langgraph.store.base at module load.

Verification

autogen   7 passed
crewai    3 passed
langgraph 6 passed

All 16 tests green locally.

[claude]

⚠️ Code review skipped — your organization's overage spend limit has been reached.

Code review is billed via overage credits. To resume reviews, an organization admin can raise the monthly limit at claude.ai/admin-settings/claude-code.

Once credits are available, push a new commit or reopen this pull request to trigger a review.