Add trace instrumentation for turn startup and response streaming (#56515)

## What changed

- Record time spent waiting for each stream event in `codex.response.receive_wait_ms` on the existing `handle_responses` span.
- Add trace spans for startup prewarm consumption, turn-start lifecycle callbacks, assistant text flushing, and response stream error handling.
- Include the turn ID and lifecycle phase on the turn-start span and the request type on the error-handling span.

GitOrigin-RevId: 7792508c830ef05b03aed01eb840ca7e6577ad0b

codex-rs/core/src/responses_retry.rs

@@ -54,6 +54,7 @@ pub(crate) struct ExhaustedResponseRetry {
 
 /// Returns `Ok(())` when the caller should retry the request loop, or the original error when
 /// it is terminal or the retry budget is exhausted.
+#[tracing::instrument(level = "trace", skip_all, fields(request = ?request))]
 pub(crate) async fn handle_response_stream_error(
     retry_state: &mut ResponsesStreamRetryState,
     max_retries: u64,

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

@@ -286,6 +286,7 @@ impl Session {
         ));
     }
 
+    #[tracing::instrument(level = "trace", skip_all)]
     pub(crate) async fn consume_startup_prewarm_for_regular_turn(
         &self,
         cancellation_token: &CancellationToken,

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

@@ -2,6 +2,7 @@ use std::collections::HashMap;
 use std::collections::HashSet;
 use std::sync::Arc;
 use std::sync::atomic::Ordering;
+use std::time::Instant;
 
 use crate::client::ModelClientSession;
 use crate::client_common::Prompt;
@@ -2418,6 +2419,7 @@ async fn emit_streamed_assistant_text_delta(
 }
 
 /// Flush buffered assistant text parser state when an assistant message item ends.
+#[instrument(level = "trace", skip_all)]
 async fn flush_assistant_text_segments_for_item(
     sess: &Session,
     turn_context: &TurnContext,
@@ -2430,6 +2432,7 @@ async fn flush_assistant_text_segments_for_item(
 }
 
 /// Flush any remaining buffered assistant text parser state at response completion.
+#[instrument(level = "trace", skip_all)]
 async fn flush_assistant_text_segments_all(
     sess: &Session,
     turn_context: &TurnContext,
@@ -2815,14 +2818,21 @@ async fn try_run_sampling_request(
             gen_ai.usage.output_tokens = field::Empty,
             codex.usage.reasoning_output_tokens = field::Empty,
             codex.usage.total_tokens = field::Empty,
+            codex.response.receive_wait_ms = field::Empty,
         );
 
+        let receive_started = Instant::now();
         let event = stream
             .next()
             .instrument(trace_span!(parent: &handle_responses, "receiving"))
             .or_cancel(&preempt)
             .or_cancel(&cancellation_token)
             .await;
+        // Split the existing iteration span without adding another span per streamed delta.
+        handle_responses.record(
+            "codex.response.receive_wait_ms",
+            receive_started.elapsed().as_secs_f64() * 1000.0,
+        );
         // Account for an already-received terminal error even if cancellation raced it.
         // Cancellation still stops ordinary stream events before they can run tools.
         let event = match event {

codex-rs/core/src/tasks/lifecycle.rs

@@ -13,6 +13,7 @@ use crate::session::session::Session;
 use crate::session::turn_context::TurnContext;
 
 impl Session {
+    #[tracing::instrument(level = "trace", skip_all, fields(turn_id = %turn_context.sub_id, phase = ?phase))]
     pub(super) async fn emit_turn_start_lifecycle(
         self: &Arc<Self>,
         turn_context: &TurnContext,