perf(merge): replace mergo pre-copy loop with reflection-free native deep merge (3.5× faster)

perf(merge): replace mergo pre-copy loop with reflection-free native deep merge (3.5× faster) #14148