Refactor image content to use a shared ImageReference type (#45543)

## What changed

Represent images in `ContentItem` and `FunctionCallOutputContentItem` with `ImageReference::Inline`, flattened to preserve the existing `image_url` wire format. Update image producers and consumers and regenerate app-server schemas and SDK artifacts.

Preserve the Python SDK's `InputImageContentItem` and `InputImageFunctionCallOutputContentItem` class names during generation.

## Testing

Add a regression test for stable Python image class names and adapt existing image tests to the shared representation.

GitOrigin-RevId: c38a780ac3314c2ac2deb3afc1b93b94b6f93fec

codex-rs/core/tests/suite/client.rs

@@ -40,6 +40,7 @@ use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::LocalShellAction;
 use codex_protocol::models::LocalShellExecAction;
 use codex_protocol::models::LocalShellStatus;
@@ -1078,7 +1079,9 @@ async fn resume_replays_legacy_js_repl_image_rollout_shapes() {
                 id: None,
                 role: "user".to_string(),
                 content: vec![ContentItem::InputImage {
-                    image_url: legacy_image_url.to_string(),
+                    image: ImageReference::Inline {
+                        image_url: legacy_image_url.to_string(),
+                    },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 }],
                 phase: None,
@@ -1218,7 +1221,9 @@ async fn resume_replays_image_tool_outputs_with_detail() {
                 namespace: None,
                 output: FunctionCallOutputPayload::from_content_items(vec![
                     FunctionCallOutputContentItem::InputImage {
-                        image_url: image_url.to_string(),
+                        image: ImageReference::Inline {
+                            image_url: image_url.to_string(),
+                        },
                         detail: Some(ImageDetail::Original),
                     },
                 ]),
@@ -1247,7 +1252,9 @@ async fn resume_replays_image_tool_outputs_with_detail() {
                 name: None,
                 output: FunctionCallOutputPayload::from_content_items(vec![
                     FunctionCallOutputContentItem::InputImage {
-                        image_url: image_url.to_string(),
+                        image: ImageReference::Inline {
+                            image_url: image_url.to_string(),
+                        },
                         detail: Some(ImageDetail::Original),
                     },
                 ]),

codex-rs/core/tests/suite/code_mode.rs

@@ -42,6 +42,7 @@ use codex_protocol::dynamic_tools::DynamicToolResponse;
 use codex_protocol::dynamic_tools::DynamicToolSpec;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::PermissionProfile;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::openai_models::InputModality;
@@ -5662,7 +5663,9 @@ async fn code_mode_node_repl_screenshots_can_be_captured_without_guardian_transc
     assert_eq!(
         evidence.images(),
         vec![ContentItem::InputImage {
-            image_url: SCREENSHOT.to_owned(),
+            image: ImageReference::Inline {
+                image_url: SCREENSHOT.to_owned()
+            },
             detail: Some(ImageDetail::Low),
         }]
     );

codex-rs/core/tests/suite/image_rollout.rs

@@ -10,6 +10,7 @@ use codex_protocol::config_types::Settings;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::PermissionProfile;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ResponseItem;
@@ -69,7 +70,10 @@ fn find_user_message_with_image(text: &str) -> Option<ResponseItem> {
 fn extract_image_url(item: &ResponseItem) -> Option<String> {
     match item {
         ResponseItem::Message { content, .. } => content.iter().find_map(|span| match span {
-            ContentItem::InputImage { image_url, .. } => Some(image_url.clone()),
+            ContentItem::InputImage {
+                image: ImageReference::Inline { image_url },
+                ..
+            } => Some(image_url.clone()),
             _ => None,
         }),
         _ => None,
@@ -178,7 +182,7 @@ async fn copy_paste_local_image_persists_rollout_request_shape() -> anyhow::Resu
                 ),
             },
             ContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             ContentItem::InputText {
@@ -269,7 +273,7 @@ async fn drag_drop_image_persists_rollout_request_shape() -> anyhow::Result<()>
         role: "user".to_string(),
         content: vec![
             ContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             ContentItem::InputText {
@@ -341,7 +345,9 @@ async fn resumed_history_only_emits_resize_notices_for_new_images() -> anyhow::R
     historical_content.insert(
         /*index*/ 0,
         ContentItem::InputImage {
-            image_url: original_image_url.clone(),
+            image: ImageReference::Inline {
+                image_url: original_image_url.clone(),
+            },
             detail: Some(ImageDetail::High),
         },
     );
@@ -480,7 +486,9 @@ async fn resumed_history_only_emits_resize_notices_for_new_images() -> anyhow::R
             ResponseInputItem::Message {
                 role: "user".to_string(),
                 content: vec![ContentItem::InputImage {
-                    image_url: original_image_url,
+                    image: ImageReference::Inline {
+                        image_url: original_image_url,
+                    },
                     detail: Some(ImageDetail::High),
                 }],
                 phase: None,

codex-rs/core/tests/suite/rmcp_client.rs

@@ -52,6 +52,7 @@ use codex_protocol::mcp_policy::McpServerRequirement;
 use codex_protocol::mcp_policy::PluginMcpRequirements;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::PermissionProfile;
 use codex_protocol::models::PermissionProfileSnapshot;
 use codex_protocol::models::ResponseItem;
@@ -1549,7 +1550,9 @@ async fn interrupt_during_mcp_startup_preserves_user_input_in_history(
     };
     assert!(
         content.contains(&ContentItem::InputImage {
-            image_url: OPENAI_PNG.to_string(),
+            image: ImageReference::Inline {
+                image_url: OPENAI_PNG.to_string()
+            },
             detail: Some(ImageDetail::Original),
         }),
         "interrupted input must use the current model's unified image budget"