[improvement] Make ConversationSummarizationTransform be able to be configured to trigger based on token counts

[dhilloulinoracle]

ConversationSummarizationTransform is currently triggered when there is a certain number of messages in the conversation, but in practice developers might be more concerned about the context size in terms of tokens.
It would therefore be useful to have the transform be configured in terms of tokens (or characters).