feat(graph): LangGraph parity shims — sync invoke + mermaid/ascii renderers

[fede-kamel]

Closes #250

Summary

Closes the 4 documented gaps between locus StateGraph and LangGraph's CompiledStateGraph:

• invoke() / run_sync() — sync wrappers around ainvoke. Refuse with a clear error when called inside a running event loop, pointing the caller at ainvoke.
• draw_mermaid() / draw_ascii() / get_mermaid() on StateGraph — so compiled.get_graph().draw_mermaid() works LangGraph-style. Both delegate to the existing locus.multiagent.visualize helpers.
• get_graph() docstring spells out the rendering chain.

run_sync and get_mermaid were already documented in docs/concepts/multi-agent/graph.md but the methods were missing — this brings the API in line with the docs.

Test plan

• TestLangGraphParityShims — 6 new tests: sync invoke, run_sync alias, running-loop guard, mermaid/ascii rendering, get_graph().draw_mermaid() chain.
• Full unit suite: uv run pytest tests/unit/ — 4794 passed.
• uv run ruff check + ruff format --check — clean.
• uv run mypy src/locus/multiagent/graph.py — no issues.
• uv run mkdocs build --strict — builds clean.

[fede-kamel]

Context — how the gaps were found

The 4 gaps closed here were caught by a side-by-side LangGraph vs Locus verifier:

📎 Gist: https://gist.github.com/fede-kamel/cb45aeac259bb995f8be23a5d7ca6965 (langgraph_vs_locus.py) — same workflow / same model, run against both frameworks. The Locus side failed on compiled.invoke(...), compiled.run_sync(...), compiled.get_mermaid(), and compiled.get_graph().draw_mermaid() — all of which now work after this PR.

Tests added

tests/unit/test_graph.py::TestLangGraphParityShims — 6 regression tests, one per gap + a running-loop guard:

Test	Covers
test_invoke_is_sync_and_runs	Gap 1 — sync compiled.invoke({...})
test_run_sync_alias	Gap 2 — run_sync matches docs
test_invoke_refuses_inside_running_loop	Safety — clear error inside live loop
test_get_mermaid_and_draw_mermaid_equal	Gap 3 — get_mermaid() matches docs
test_get_graph_chain_to_draw_mermaid	Gap 4 — compiled.get_graph().draw_mermaid() chain
test_draw_ascii_runs	Renderer parity — .draw_ascii() on StateGraph

All 6 pass locally and in CI (4794 unit tests green).