Set default model to GPT-3.5

[heatherlogan-scottlogic]

GPT-4 is 20 times more expensive than 3.5, we have been included in some emails re. costs so good idea to switch back for now. Also It seems like 3.5 can be easier to break than 4, so when we develop the new story / instructions/system roles, makes sense to do so with 3.5 in mind.
GPT-3.5 will definitely be adequate for document QA models and LLM evaluation models.
GPT-4 is currently used as default for the main conversation bot - we can replace this with GPT-3.5 and allow used to select GPT-4 in the sandbox.

This ticket

• use GPT-3.5 as document QA and LLM evaluation bots
• use GPT-3.5 as default for main conversation

[asaha-scottlogic]

looks good, defaulted to 3.5 turbo