Implement spend_cpu_cycles in assembly

The compiler was unrolling it previously, so we're doing assembly now for consistency.

config.toml

@@ -29,6 +29,6 @@ keepalive_interval_seconds = 30
 [benchmark]
 # CPU-intensive loop iterations for /ok requests in resolve stage
 # 0 = health check only (no CPU work)
-# 7000 = default benchmark load (650ns CPU work, 1M req/s)
+# 4000 = default benchmark load (740ns CPU work, 1M req/s)
 # Higher values = more CPU stress testing
 ok_resolve_iterations = 0