Trace MCP runtime refresh coordination (#39667)

## What changed

Add named tracing spans for MCP dirty-state refreshes and time spent waiting
to acquire the refresh semaphore.

GitOrigin-RevId: 1dbb3572d2061581beb83b3de098d6ec91791104

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

@@ -154,6 +154,7 @@ impl Session {
     }
 
     /// Publishes changed MCP state, waiting for any refresh already in progress.
+    #[tracing::instrument(name = "mcp.runtime.refresh_if_dirty", skip_all)]
     pub(crate) async fn refresh_mcp_if_dirty(self: &Arc<Self>) {
         let Ok(_refresh) = self.mcp_refresh.acquire().await else {
             error!("MCP runtime refresh semaphore closed");

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

@@ -30,6 +30,7 @@ impl McpRefresh {
         self.pending.swap(false, Ordering::AcqRel)
     }
 
+    #[tracing::instrument(name = "mcp.runtime.refresh_wait", skip_all)]
     pub(super) async fn acquire(&self) -> Result<SemaphorePermit<'_>, AcquireError> {
         self.gate.acquire().await
     }