feat: consult Approvisionnement

One entry in the registry and a realm grant, which was the claim the design made
about adding a specialist. Nothing in the graph changed.

What the two specialists have in common, from the orchestrator's side, is
nothing: one is PydanticAI over MCP, the other n8n workflows behind an adapter.
It cannot tell them apart, which is what the protocol is for.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

src/assistant_agent/specialists.py

@@ -50,10 +50,26 @@ CATALOGUE = Specialist(
 )
 
 
-#: Everything the design foresees, minus what is not built. Ventes, Support,
-#: Approvisionnement and Recouvrement belong here as they arrive; the graph
-#: needs no change for them, only this list and a realm grant.
-SPECIALISTS: tuple[Specialist, ...] = (CATALOGUE,)
+APPROVISIONNEMENT = Specialist(
+    name="approvisionnement",
+    url="http://approvisionnement-agent.agents.svc.cluster.local:8080/",
+    audience_scope="classic-models-agent-approvisionnement-audience",
+    serves="supply: whether a reference can be promised, from stock on hand",
+    decides=(
+        "whether a product can be promised to a customer, and why not when it cannot",
+        "whether a reference is close to needing a reorder",
+    ),
+)
+
+
+#: Everything the design foresees, minus what is not built. Ventes, Support and
+#: Recouvrement belong here as they arrive; the graph needs no change for them,
+#: only this list and a realm grant.
+#:
+#: Note what these two have in common from here: nothing. One is PydanticAI over
+#: MCP, the other n8n behind an adapter. The orchestrator cannot tell, and that
+#: is the point of the protocol.
+SPECIALISTS: tuple[Specialist, ...] = (CATALOGUE, APPROVISIONNEMENT)
 
 
 def by_name(name: str) -> Specialist | None: