Topic deduplication/merging now separated from summarisation. Gradio upgrade

Moved spaces GPU calls back to main functions as otherwise it doesn't seem to work correctly

Trying to move calls to @spaces.GPU to specific gemma calls to use the local model more efficiently

Enhanced local model support by adding model loading functionality in chatfuncs.py and updating llm_api_call.py to utilize local models for topic extraction and summarization. Improved model path handling and ensured compatibility with GPU configurations.

Refactored call_llama_cpp_model function to include model parameter in chatfuncs.py and updated import statements in llm_api_call.py to reflect this change.

Moved model load to chatfuncs submodule to hopefully avoid gpu run issues

Adding some compatibility with Zero GPU spaces

Added support for using local models (specifically Gemma 2b) for topic extraction and summary. Generally improved output format safeguards.