Fix the ordering of the middlewares

This was causing the number of event processors to constantly grow
with each request, making the server use up more memory and CPU over
time.

crates/cli/src/server.rs

@@ -293,6 +293,7 @@ pub fn build_router(
     router = router.fallback(mas_handlers::fallback);
 
     router
+        .layer(axum::middleware::from_fn(log_response_middleware))
         .layer(
             InFlightCounterLayer::new("http.server.active_requests").on_request((
                 name.map(|name| KeyValue::new(MAS_LISTENER_NAME, name.to_owned())),
@@ -318,12 +319,16 @@ pub fn build_router(
                 span.record("otel.status_code", "OK");
             }),
         )
-        .layer(axum::middleware::from_fn(log_response_middleware))
         .layer(mas_context::LogContextLayer::new(|req| {
             otel_http_method(req).into()
         }))
-        .layer(NewSentryLayer::new_from_top())
+        // Careful about the order here: the `NewSentryLayer` must be around the
+        // `SentryHttpLayer`. axum makes new layers wrap the existing ones,
+        // which is the other way around compared to `tower::ServiceBuilder`.
+        // So even if the Sentry docs has an example that does
+        // 'NewSentryHttpLayer then SentryHttpLayer', we must do the opposite.
         .layer(SentryHttpLayer::with_transaction())
+        .layer(NewSentryLayer::new_from_top())
         .with_state(state)
 }