fix(openai): tolerate prompt_cache_retention drift in streaming
#36925
+109
−15
prompt_cache_retention drift in streaming
#36925