use plano-orchestrator for LLM routing, remove arch-router

Replace RouterService/RouterModelV1 (arch-router prompt) with
OrchestratorService/OrchestratorModelV1 (plano-orchestrator prompt)
for LLM routing. This ensures the correct system prompt is used when
llm_routing_model points at a Plano-Orchestrator model.

- Extend OrchestratorService with session caching, ModelMetricsService,
  top-level routing preferences, and determine_route() for LLM routing
- Delete RouterService, RouterModel trait, RouterModelV1, and
  ARCH_ROUTER_V1_SYSTEM_PROMPT
- Unify defaults to Plano-Orchestrator / plano-orchestrator
- Update CLI config generator, demos, docs, and config schema

Made-with: Cursor

crates/common/src/configuration.rs

@@ -729,13 +729,6 @@ mod test {
                 internal: None,
                 ..Default::default()
             },
-            LlmProvider {
-                name: "arch-router".to_string(),
-                provider_interface: LlmProviderType::Plano,
-                model: Some("Arch-Router".to_string()),
-                internal: Some(true),
-                ..Default::default()
-            },
             LlmProvider {
                 name: "plano-orchestrator".to_string(),
                 provider_interface: LlmProviderType::Plano,
@@ -747,13 +740,10 @@ mod test {
 
         let models = providers.into_models();
 
-        // Should only have 1 model: openai-gpt4
         assert_eq!(models.data.len(), 1);
 
-        // Verify internal models are excluded from /v1/models
         let model_ids: Vec<String> = models.data.iter().map(|m| m.id.clone()).collect();
         assert!(model_ids.contains(&"openai-gpt4".to_string()));
-        assert!(!model_ids.contains(&"arch-router".to_string()));
         assert!(!model_ids.contains(&"plano-orchestrator".to_string()));
     }
 }