fix: race condition in hop-by-hop routing where state saved after send

The operation state was being pushed AFTER sending the forwarded message.
In cases where the remote peer responded quickly, the response could arrive
before the state was saved, causing load_or_init to fail to find the
operation (returning OpNotPresent).

This caused flaky failures in test_put_contract_three_hop_returns_response
in CI environments where timing was faster than local testing.

Fix: Push state to op_manager BEFORE sending the message to ensure the
state is always available when the response arrives.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: sanity

crates/core/src/operations/mod.rs

@@ -169,8 +169,11 @@ where
                 tracing::debug!(%id, "operation in progress");
                 if let Some(target) = next_hop {
                     tracing::debug!(%id, ?target, "sending updated op state");
-                    network_bridge.send(target, msg).await?;
+                    // IMPORTANT: Push state BEFORE sending message to avoid race condition.
+                    // If we send first, a fast response might arrive before the state is saved,
+                    // causing load_or_init to fail to find the operation.
                     op_manager.push(id, updated_state).await?;
+                    network_bridge.send(target, msg).await?;
                 } else {
                     tracing::debug!(%id, "queueing op state for local processing");
                     debug_assert!(