feat(infra): add ollama to on-demand sidecar lifecycle

Ollama now starts on API request and stops after 5min idle, saving
~5-6GB RAM when not in use. Part of server-wide resource caps rollout.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

server/index.ts

@@ -1969,6 +1969,7 @@ app.post("/api/prompt", async (c) => {
 
 	if (OLLAMA_MODELS[model]) {
 		try {
+			await ensureSidecar("ollama");
 			const ollamaRes = await fetch(`${OLLAMA_URL}/api/chat`, {
 				method: "POST",
 				headers: { "Content-Type": "application/json" },
@@ -1989,6 +1990,7 @@ app.post("/api/prompt", async (c) => {
 			}
 
 			const data = await ollamaRes.json();
+			markSidecarUsed("ollama");
 			return c.json({ content: data.message?.content || "" });
 		} catch (e: any) {
 			console.error("[prompt] Ollama unreachable:", e.message);

server/sidecar-manager.ts

@@ -39,6 +39,12 @@ const SIDECARS: Record<string, SidecarConfig> = {
 		port: 8810,
 		healthTimeout: 15_000,
 	},
+	ollama: {
+		container: "ollama",
+		host: "ollama",
+		port: 11434,
+		healthTimeout: 30_000,
+	},
 };
 
 const lastUsed = new Map<string, number>();