Fix token limit

[Czechh]

Fixes token limit issues in incremental summarization. Now the algorithm is the following:

• We take window_size / 2 messages to summarize.
• We verify the token count message by message until the max token size limit is reached.
• Then we summarize that subset of messages.
• And repeat with the rest of the list.
• At the end we keep track of the total number of tokens used for the process.