Refactor prompt construction and token management in LLM inference process, enhancing context handling and logging for code prompts

Add new model presets for GPT-5.2, GPT-5.1, and Claude 4.5 series, while removing outdated GPT-4 and Claude 3 models for improved model selection.

Refactor LLM parameter handling to improve compatibility with GPT-5 and GPT-OSS models

Add drop_params option to LLM initialization and adjust parameter handling for model compatibility

disable local models for now

inital commit