refactor(relay): trim plugin-only surface

Signed-off-by: Bryan Bednarski <bbednarski@nvidia.com>

Author: bbednarski9

crates/switchyard-nemo-relay-plugin/src/client.rs

@@ -73,11 +73,6 @@ impl TargetClient {
         metadata.http_headers = None;
         request
     }
-
-    #[cfg(test)]
-    fn provider_model(&self) -> &str {
-        &self.provider_model
-    }
 }
 
 #[async_trait]
@@ -161,12 +156,6 @@ mod tests {
         );
     }
 
-    #[test]
-    fn semantic_selection_does_not_replace_the_provider_model() {
-        let client = client(WireFormat::OpenAiChat);
-        assert_eq!(client.provider_model(), "provider/model");
-    }
-
     #[test]
     fn only_anthropic_targets_advertise_count_tokens() {
         assert!(client(WireFormat::AnthropicMessages).supports_count_tokens());

crates/switchyard-nemo-relay-plugin/src/config.rs

@@ -162,8 +162,6 @@ enum AlgorithmConfig {
         classifier_target: String,
         weak_target: String,
         strong_target: String,
-        #[serde(default)]
-        escalation: Option<serde_json::Value>,
         #[serde(flatten)]
         config: TaskClassifierConfig,
     },
@@ -296,15 +294,8 @@ impl SwitchyardConfig {
                 classifier_target,
                 weak_target,
                 strong_target,
-                escalation,
                 config,
             } => {
-                if escalation.is_some() {
-                    return Err(
-                        "llm_classifier escalation mode is not supported by this plugin version"
-                            .into(),
-                    );
-                }
                 let classifier_binding = self.targets.get(classifier_target).ok_or_else(|| {
                     format!("algorithm target {classifier_target:?} is not configured")
                 })?;
@@ -357,7 +348,7 @@ fn validate_dispatch_url(
     }
 
     // The current switchyard-llm-client accepts provider base URLs and complete
-    // canonical endpoints. Reject a custom terminal route here instead of
+    // canonical endpoints. Reject a custom terminal route to avoid
     // allowing Backend::url() to append another provider suffix silently.
     let expected_suffix = match protocol {
         WireFormat::OpenAiChat => "/chat/completions",
@@ -686,7 +677,6 @@ mod tests {
             classifier_target: "anthropic".into(),
             weak_target: "responses".into(),
             strong_target: "chat".into(),
-            escalation: None,
             config: TaskClassifierConfig {
                 base_threshold: 0.5,
                 ..Default::default()
@@ -743,7 +733,6 @@ mod tests {
             classifier_target: "chat".into(),
             weak_target: "responses".into(),
             strong_target: "anthropic".into(),
-            escalation: None,
             config: TaskClassifierConfig {
                 base_threshold: 1.1,
                 ..Default::default()

crates/switchyard-nemo-relay-plugin/src/runtime.rs

@@ -115,9 +115,15 @@ impl SwitchyardRuntime {
                         failure_mark_data(attempt, &failure),
                         &metadata,
                     );
-                    return self
-                        .fallback_buffered(inbound, request, marks, &metadata)
-                        .await;
+                    let response = self
+                        .fallback_response(inbound, request, marks, &metadata)
+                        .await?;
+                    let LlmResponse::Agg(response) = response.llm_response else {
+                        return Err(
+                            "trusted fallback returned a stream for a buffered request".into()
+                        );
+                    };
+                    return translation::encode_response(&self.translation, inbound, &response);
                 }
             }
         }
@@ -322,22 +328,6 @@ impl SwitchyardRuntime {
         call.respond(result)
     }
 
-    async fn fallback_buffered(
-        &self,
-        inbound: WireFormat,
-        request: Request,
-        marks: &mut Vec<RoutingMark>,
-        metadata: &Json,
-    ) -> Result<Json, String> {
-        let response = self
-            .fallback_response(inbound, request, marks, metadata)
-            .await?;
-        let LlmResponse::Agg(response) = response.llm_response else {
-            return Err("trusted fallback returned a stream for a buffered request".into());
-        };
-        translation::encode_response(&self.translation, inbound, &response)
-    }
-
     async fn fallback_response(
         &self,
         inbound: WireFormat,