Make `LLMClient.executeStreaming` expose more information

[OKatrych]

Currently, when using LLMClient.executeStreaming, it returns only a flow of content chunks Flow<String>. This is sufficient for simple cases, but not enough for production applications that require displaying more information to the user.
Things that I missed:

• finish reason (in the same way as LLMClient.execute has), we don't know a reason why streaming suddenly stopped (some error, or it was the expected ending)
• All the errors that happen during the stream are skipped. Take, for example, AnthropicLLMClient:

event
    .takeIf { it.event == "content_block_delta" } // "error" event is skipped
    ?.data?.trim()?.let { json.decodeFromString<AnthropicStreamResponse>(it) }
    ?.delta?.text?.let { emit(it) }

• I think it was already mentioned here, we don't save the message ID received from the provider

Currently, in my project, I use this class to represent the stream result from any provider:

sealed interface StreamResult {

    data class Chunk(
        val id: String,
        val content: String,
    ) : StreamResult

    data class Finished(
        val id: String,
        val reason: FinishReason?,
    ) : StreamResult

    data class Error(val error: String?) : StreamResult
}

It would be nice to see something similar in Koog.