feat: support cached token usage

[wellenzheng]

support cached token usage

while glm-4.5 series models are supporting, we have the new usage strategy cached_token, we are hoping to add the cached_token support in the LLMUsage.

glm-4.5 usage sample

"usage": {
"completion_tokens": 139,
"prompt_tokens": 15427,
"prompt_tokens_details": {
"cached_tokens": 170
},
"total_tokens": 15566
}