Attribute turn metrics to the models used during the turn (#44656)

## Why

Turn completion metrics used session-level telemetry, which could label usage with the wrong model after a model switch. A turn can also include compaction and responses from different models, so a single token total cannot attribute usage correctly.

## What changed

- Accumulate token usage by model using each response's resolved step settings, and emit one histogram sample per model and token type per turn.
- Preserve zero-valued samples when no usage is reported, without adding a sample for an unused selected model when only compaction ran.
- Use turn telemetry for tool-call, memory, network-proxy, and running-process metrics.

## Testing

Add regression tests for model switches, step-level model changes, compaction followed by a response or an aborted turn, and turns without responses. Verify repeated requests aggregate into one sample per model and earlier session usage is excluded.

GitOrigin-RevId: 03ab1dcde7da0b16b3a500a9a05e4c86583ee585

codex-rs/core/src/session/mod.rs

@@ -4533,7 +4533,7 @@ impl Session {
         token_usage: Option<&TokenUsage>,
     ) -> CodexResult<()> {
         let result = self
-            .record_token_usage_info(turn_context, token_usage)
+            .record_token_usage_info(turn_context, &turn_context.initial_settings, token_usage)
             .await;
         self.send_token_count_event(turn_context).await;
         result
@@ -4576,6 +4576,7 @@ impl Session {
     pub(crate) async fn record_token_usage_info(
         &self,
         turn_context: &TurnContext,
+        settings: &ResolvedStepSettings,
         token_usage: Option<&TokenUsage>,
     ) -> CodexResult<()> {
         if let Some(token_usage) = token_usage {
@@ -4591,6 +4592,17 @@ impl Session {
                 }
                 state.token_info()
             };
+            let turn_state = self
+                .input_queue
+                .turn_state_for_sub_id(&self.active_turn, &turn_context.sub_id)
+                .await;
+            if let Some(turn_state) = turn_state {
+                turn_state.lock().await.token_usage_by_model.record(
+                    settings.selected_collaboration_mode().model(),
+                    settings.telemetry(&turn_context.session_telemetry),
+                    token_usage,
+                );
+            }
             let budget_result = self.record_rollout_budget_usage(token_usage);
             if let Some(token_info) = token_info.as_ref() {
                 for contributor in self.services.extensions.token_usage_contributors() {

codex-rs/core/src/session/tests.rs

@@ -3121,11 +3121,19 @@ async fn record_token_usage_info_notifies_extension_contributors() {
     };
 
     session
-        .record_token_usage_info(&turn_context, Some(&first_usage))
+        .record_token_usage_info(
+            &turn_context,
+            &turn_context.initial_settings,
+            Some(&first_usage),
+        )
         .await
         .expect("first usage should be recorded");
     session
-        .record_token_usage_info(&turn_context, Some(&second_usage))
+        .record_token_usage_info(
+            &turn_context,
+            &turn_context.initial_settings,
+            Some(&second_usage),
+        )
         .await
         .expect("second usage should be recorded");
 

codex-rs/core/src/session/turn.rs

@@ -2811,7 +2811,11 @@ async fn try_run_sampling_request(
                 )
                 .await;
                 let budget_result = sess
-                    .record_token_usage_info(&turn_context, token_usage.as_ref())
+                    .record_token_usage_info(
+                        &turn_context,
+                        &step_context.settings,
+                        token_usage.as_ref(),
+                    )
                     .await;
                 should_emit_token_count = true;
                 should_emit_turn_diff = true;