fix(core) Preserve base_instructions in SessionMeta (#9427)

## Summary
This PR consolidates base_instructions onto SessionMeta /
SessionConfiguration, so we ensure `base_instructions` is set once per
session and should be (mostly) immutable, unless:
- overridden by config on resume / fork
- sub-agent tasks, like review or collab


In a future PR, we should convert all references to `base_instructions`
to consistently used the typed struct, so it's less likely that we put
other strings there. See #9423. However, this PR is already quite
complex, so I'm deferring that to a follow-up.

## Testing
- [x] Added a resume test to assert that instructions are preserved. In
particular, `resume_switches_models_preserves_base_instructions` fails
against main.

Existing test coverage thats assert base instructions are preserved
across multiple requests in a session:
- Manual compact keeps baseline instructions:
core/tests/suite/compact.rs:199
- Auto-compact keeps baseline instructions:
core/tests/suite/compact.rs:1142
- Prompt caching reuses the same instructions across two requests:
core/tests/suite/prompt_caching.rs:150 and
core/tests/suite/prompt_caching.rs:157
- Prompt caching with explicit expected string across two requests:
core/tests/suite/prompt_caching.rs:213 and
core/tests/suite/prompt_caching.rs:222
- Resume with model switch keeps original instructions:
core/tests/suite/resume.rs:136
- Compact/resume/fork uses request 0 instructions for later expected
payloads: core/tests/suite/compact_resume_fork.rs:215

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

@@ -388,9 +388,10 @@ async fn remote_models_apply_remote_base_instructions() -> Result<()> {
 
     wait_for_event(&codex, |event| matches!(event, EventMsg::TurnComplete(_))).await;
 
+    let base_model_info = models_manager.get_model_info("gpt-5.1", &config).await;
     let body = response_mock.single_request().body_json();
     let instructions = body["instructions"].as_str().unwrap();
-    assert_eq!(instructions, remote_base);
+    assert_eq!(instructions, base_model_info.base_instructions);
 
     Ok(())
 }

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

@@ -131,3 +131,78 @@ async fn resume_includes_initial_messages_from_reasoning_events() -> Result<()>
 
     Ok(())
 }
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn resume_switches_models_preserves_base_instructions() -> Result<()> {
+    skip_if_no_network!(Ok(()));
+
+    let server = start_mock_server().await;
+    let mut builder = test_codex().with_config(|config| {
+        config.model = Some("gpt-5.2".to_string());
+    });
+    let initial = builder.build(&server).await?;
+    let codex = Arc::clone(&initial.codex);
+    let home = initial.home.clone();
+    let rollout_path = initial.session_configured.rollout_path.clone();
+
+    let initial_sse = sse(vec![
+        ev_response_created("resp-initial"),
+        ev_assistant_message("msg-1", "Completed first turn"),
+        ev_completed("resp-initial"),
+    ]);
+    let initial_mock = mount_sse_once(&server, initial_sse).await;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![UserInput::Text {
+                text: "Record initial instructions".into(),
+                text_elements: Vec::new(),
+            }],
+            final_output_json_schema: None,
+        })
+        .await?;
+    wait_for_event(&codex, |event| matches!(event, EventMsg::TurnComplete(_))).await;
+
+    let initial_body = initial_mock.single_request().body_json();
+    let initial_instructions = initial_body
+        .get("instructions")
+        .and_then(|v| v.as_str())
+        .unwrap_or_default()
+        .to_string();
+
+    let resumed_sse = sse(vec![
+        ev_response_created("resp-resume"),
+        ev_assistant_message("msg-2", "Resumed turn"),
+        ev_completed("resp-resume"),
+    ]);
+    let resumed_mock = mount_sse_once(&server, resumed_sse).await;
+
+    let mut resume_builder = test_codex().with_config(|config| {
+        config.model = Some("gpt-5.2-codex".to_string());
+    });
+    let resumed = resume_builder.resume(&server, home, rollout_path).await?;
+    resumed
+        .codex
+        .submit(Op::UserInput {
+            items: vec![UserInput::Text {
+                text: "Resume with different model".into(),
+                text_elements: Vec::new(),
+            }],
+            final_output_json_schema: None,
+        })
+        .await?;
+    wait_for_event(&resumed.codex, |event| {
+        matches!(event, EventMsg::TurnComplete(_))
+    })
+    .await;
+
+    let resumed_body = resumed_mock.single_request().body_json();
+    let resumed_instructions = resumed_body
+        .get("instructions")
+        .and_then(|v| v.as_str())
+        .unwrap_or_default()
+        .to_string();
+    assert_eq!(resumed_instructions, initial_instructions);
+
+    Ok(())
+}

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

@@ -28,7 +28,6 @@ fn resume_history(
         model: previous_model.to_string(),
         effort: config.model_reasoning_effort,
         summary: config.model_reasoning_summary,
-        base_instructions: None,
         user_instructions: None,
         developer_instructions: None,
         final_output_json_schema: None,