Avoid sending the user comments twice by directly evaluating them from the list with all the comments

[gentlementlegen]

          Yeah that's why we could leave `allComments` so it makes more sense to the LLM but remove `userComments` and specify that the LLM should evaluate comments of user `x`. 

If I understand correctly comments include id and author so the LLM would know which comments to evaluate, but I might be wrong maybe it will miss comments

Originally posted by @whilefoo in #225 (comment)

To evaluate user comments, we currently send the whole list of the comments under the issue, then the list of user comments. This means we send twice the same content, since the user comments are already in the comment list. Instead, we should tell in the prompt what user we are evaluating to extract them during the LLM evaluation, which would greatly reduce the amount of tokens used, allowing us to send more data.

We could also look into prompt caching, although it might be specific to OpenAI API.

[ubiquity-os-beta]

Note

The following contributors may be suitable for this task:

Keyrxng

79% Match ubiquity-os/plugins-wishlist#29

jaykayudo

78% Match ubiquity-os-marketplace/text-conversation-rewards#147

gentlementlegen

77% Match ubiquity-os-marketplace/text-conversation-rewards#103

[0x4007]

@shiv810 please edit the spec for any clarifications as it seems you have the best understanding of the cache mechanism.