Update our model selection to include GPT-5, and use it as part of the table generation

[PClmnt]

Description

This PR updates our model selector to have the three new GPT-5 models and some accompanying updates around parameters to support this.

• Removed the old GPT turbo models from the options. I can't really see a use case where you would ever want to use these. And gpt-4-turbo is more expensive than even gpt-5 (per token, it might generate less i suppose).
• Update the OpenAI library to 5.X.X

Launchcontrol

• Allow selection of GPT-5 in the AI model settings

[qa-wolf]

QA Wolf here! As you write new code it's important that your test coverage is keeping up.
Click here to request test coverage for this PR!

[mike12345567]

Amazing job working this out - seems like it was a real pain! Just to check, does this mean we are now using the unidici approach for our mock model responses, but nock everywhere else?
If there is a specific error you remember seeing, could be worth documenting somewhere what to do if you need to mock a builtin node fetch.

[PClmnt]

Amazing job working this out - seems like it was a real pain! Just to check, does this mean we are now using the unidici approach for our mock model responses, but nock everywhere else? If there is a specific error you remember seeing, could be worth documenting somewhere what to do if you need to mock a builtin node fetch.

That's right, yea. I know it's not ideal but migrating everything over from Nock seemed like being more of a pain. And at least now we have a reasonable concept of how to do this, as we will certainly start to see it as more libs start to move towards undici / internal fetch over node-fetch