in tokenizer.ts

import { countTokens } from "gpt-tokenizer/model/gpt-4o"
countTokens will throw when content is null

note: message.content can be null in assistant tool call message