Add custom default Retriever/TextSplitter, update deprecated OpenAI model

[stefanorosanelli]

This PR updates and improves general Brevia settings:

• gpt-4o-mini is now used as defatul OpenAI model instead of the deprecated gpt-3.5-turbo*
• a new general setting text_splitter variable to activate a custom default TextSplitter via a JSON configuration => enabled using a newTEXT_SPLITTER env variabile
• a new general setting qa_retriever variable to activate a custom default Retriever via a JSON configuration to be used in /chat endpoing => enabled using a new QA_RETRIVER env variabile
• (fix) chunk_size and chunk_overlap in a collection metadata config are now correctly handled:
  • if present they override general settings
  • same params specified in text_splitter conf will in turn override the previous ones