refactor: cortex talks to neurons instead of mistral.rs directly

Replace NodeConfig (static vram_mb, pinned) with NeuronEndpoint.
Hardware discovery and model pinning now come from neuron API and
models.toml catalogue respectively.

- config.rs: nodes -> neurons, add models_config path
- catalogue.rs: ModelProfile with pinned_on, ModelCatalogue
- poller.rs: poll neuron GET /models (ModelInfo format)
- router.rs: resolve inference endpoint via neuron GET /models/{id}/endpoint
- evictor.rs: call neuron POST /models/unload
- node.rs: remove vram_mb, pinned fields (come from discovery/catalogue)
- All 22 gateway tests updated to mock neuron API
- Remove MistralModelsResponse, ModelLifecycleRequest (no longer needed)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/cortex-gateway/tests/metrics.rs

@@ -4,21 +4,17 @@ use serde_json::json;
 
 #[tokio::test]
 async fn test_metrics_emitted_after_proxy() {
-    // Install a test recorder (no HTTP listener, renders to string).
-    // This sets the global recorder, so only one test can do this.
     let handle = cortex_gateway::metrics::install_test_recorder().expect("recorder should install");
 
-    let mock_url = common::spawn_mock_backend().await;
+    let mock_url = common::spawn_mock_neuron().await;
     let gw_url = common::spawn_gateway(&mock_url).await;
 
-    // Verify no request metrics yet.
     let before = handle.render();
     assert!(
         !before.contains("cortex_requests_total"),
         "no request metrics before any requests"
     );
 
-    // Make a successful request.
     let client = reqwest::Client::new();
     let resp = client
         .post(format!("{gw_url}/v1/chat/completions"))
@@ -31,10 +27,8 @@ async fn test_metrics_emitted_after_proxy() {
         .await
         .expect("request should succeed");
     assert_eq!(resp.status(), 200);
-    // Consume the response body to ensure the proxy completes.
     let _body: serde_json::Value = resp.json().await.unwrap();
 
-    // Check metrics were emitted.
     let after = handle.render();
 
     assert!(
@@ -45,7 +39,6 @@ async fn test_metrics_emitted_after_proxy() {
         after.contains("cortex_request_duration_seconds"),
         "cortex_request_duration_seconds should be present.\nMetrics:\n{after}"
     );
-    // Should NOT have error or cold start counters for this request.
     assert!(
         !after.contains("cortex_request_errors_total"),
         "no errors expected for a successful request"