Stabilize Guardian client cache key handling (#24891)

Split from the Guardian prompt cache key change. This PR only updates
codex-rs/core/src/client.rs. Validation was not run per request; this
branch is expected to rely on the companion split PRs.

codex-rs/core/src/client.rs

@@ -218,6 +218,7 @@ impl RequestRouteTelemetry {
 #[derive(Debug, Clone)]
 pub struct ModelClient {
     state: Arc<ModelClientState>,
+    prompt_cache_key_override: Option<String>,
 }
 
 /// A turn-scoped streaming session created from a [`ModelClient`].
@@ -352,9 +353,24 @@ impl ModelClient {
                 disable_websockets: AtomicBool::new(false),
                 cached_websocket_session: StdMutex::new(WebsocketSession::default()),
             }),
+            prompt_cache_key_override: None,
         }
     }
 
+    pub(crate) fn with_prompt_cache_key_override(
+        mut self,
+        prompt_cache_key_override: Option<String>,
+    ) -> Self {
+        self.prompt_cache_key_override = prompt_cache_key_override;
+        self
+    }
+
+    fn prompt_cache_key(&self) -> String {
+        self.prompt_cache_key_override
+            .clone()
+            .unwrap_or_else(|| self.state.thread_id.to_string())
+    }
+
     /// Creates a fresh turn-scoped streaming session.
     ///
     /// This constructor does not perform network I/O itself; the session opens a websocket lazily
@@ -749,7 +765,7 @@ impl ModelClient {
             &prompt.output_schema,
             prompt.output_schema_strict,
         );
-        let prompt_cache_key = Some(self.state.thread_id.to_string());
+        let prompt_cache_key = Some(self.prompt_cache_key());
         let service_tier = model_info.service_tier_for_request(service_tier);
         let request = ResponsesApiRequest {
             model: model_info.slug.clone(),