feat(codex): 按操作语义路由 Responses V2 压缩

(cherry picked from commit 2fc604e047)

apps/aether-gateway/src/ai_serving/planner/candidate_materialization.rs

@@ -706,6 +706,7 @@ pub(crate) async fn build_lazy_requested_model_execution_candidate_attempt_sourc
     trace_id: &str,
     client_api_format: &str,
     requested_model: &str,
+    request_operation: Option<&str>,
     require_streaming: bool,
     auth_snapshot: &GatewayAuthApiKeySnapshot,
     client_session_affinity: Option<&ClientSessionAffinity>,
@@ -734,6 +735,7 @@ where
         model_directive_policy,
         client_api_format,
         requested_model,
+        request_operation,
         require_streaming,
         required_capabilities,
         auth_snapshot,
@@ -1141,6 +1143,7 @@ async fn resolve_priority_candidate_page_with_cache(
 
     let key = CandidateResolvedPageCacheKey::new(
         &cursor.requested_model,
+        cursor.page_cursor.resolved_page_cache_request_operation(),
         &cursor.client_api_format,
         true,
         &cursor.auth_snapshot,
@@ -2190,6 +2193,7 @@ mod tests {
             &model_directive_policy,
             "openai:chat",
             "gpt-5",
+            None,
             true,
             None,
             &auth_snapshot,
@@ -2245,6 +2249,7 @@ mod tests {
             &model_directive_policy,
             "openai:chat",
             "gpt-5",
+            None,
             true,
             None,
             &auth_snapshot,
@@ -2282,6 +2287,7 @@ mod tests {
             &model_directive_policy,
             "openai:chat",
             "gpt-5",
+            None,
             true,
             None,
             &auth_snapshot,