Memory limit by tokens

[javipacheco]

This PR changes the strategy for getting the messages in the conversation.

Now, we have a limit and we get the last messages for this limit. In this PR we are adding a new field to Messages, the approxTokens. In this field, we have the approximate token for every message.

When we call the memories function in the VectorStore, now we have to pass the number of token that we want to receive for the last messages in the conversation

suspend fun memories(conversationId: ConversationId, limitTokens: Int): List<Memory>

This optimises the way to get the message from the VectorStore and we are sending only the necessary tokens to OpenAI

This PR changes all of the implementations of VectorStore: LocalVectorStore, CombinedVectorStore , PGVectorStore and Lucene

[javipacheco]

@raulraja @Montagon I have tested the xef-server on my local machine and the new approx_tokens field in xef_memory table is created. Everything works as expected

[raulraja]

thank you @javipacheco !