O1 parsing code replace fix

[LessComplexity]

O1 models sometimes return the following format when marking lines to replaces:

**Replace Lines**: 3-23
**Replace Lines: 2-23**
**Replace Lines**: `3-23`

This format doesn't allow completion to work correctly, so I've added a fix that clears the current line to parse the "Replace Lines" correctly on O1 models too.
Since the new code only cleans the ** and ` characters -> it still works as expected with Claude and all other models.

[yetone]

‌‌‌Hello, avante.nvim has refactored the prompt and is now using aider's prompt. It will no longer rely on line numbers returned by LLM, so it seems this issue won't occur anymore. You can try updating and testing it.

#671

[LessComplexity]

‌‌‌Hello, avante.nvim has refactored the prompt and is now using aider's prompt. It will no longer rely on line numbers returned by LLM, so it seems this issue won't occur anymore. You can try updating and testing it.

#671

Awesome I saw the new change and it's great architecture wise :)
Problem is, that it didn't take into address correctly the situation when the messages are not streaming. This works on every other model than O1.
Basically we need to add something a bit different for non-streaming messages from the models since from my tests it doesn't work at all without streaming (AKA O1 models).

To fix this, I think the best option is either to change the "on_chunk" function to handle non-streaming requests (the function fails to work in the openai "parse_response_without_stream" function), or to separate the searching/replacing from the "on_chunk" itself, and allowing it to work both during streaming and non streaming - in the same place in the code.

I will try to work on one of the options now to let the O1 models I use work with it fluently too, will update on progress :)

[LessComplexity]

@yetone
I've made it work!
My comment above is irrelevant, the code actually does suppose to work with the O1 model, the thing that made it not work is the use of this part in the code:

      if is_first_chunk then
        is_first_chunk = false
        self:update_content(content_prefix .. chunk, { stream = false, scroll = true })
        return

Since O1 models give a result it one chunk, this code was just returning the content without replacing the relevant lines for parsing. I've basically removed this as this doesn't affect the Claude and other streaming models and provides the ability to work with O1 models as they are meant too.

[LessComplexity]

I've fixed the new version of the code to work with O1 models.
The fix is pretty similar to before - use transformed content instead of chunk on first_chunk, cause no-stream responses (like those from O1) work in a single chunk either way.

The fix doesn't interfere with other functionality of different models/providers from my tests :)