feat: Update OpenAIAnswerGenerator defaults and with learnings from PromptNode

[sjrl]

Related Issues

• fixes N/A

Proposed Changes:

Breaking Change: Update to Default Values

• Updated the default values for presence_penalty and frequency_penalty to the recommended values by OpenAI to avoid repetitive text.

Reasonable values for the penalty coefficients are around 0.1 to 1 if the aim is to just reduce repetitive samples somewhat.

• Previously we had the values set to -2 (now both are 0.1), which greatly increased the chance of repetitive text being generated and is what caused the occurrence of answers ending with repetitive commas or periods.
• Updated the default model to text-davinci-003 since it returns much higher quality of results than the text-curie-001 model
• Update the max_tokens to 50 since davinci-003 has a much bigger max seq length of 4000 tokens and the generated answers are often quite verbose

Additional Features (non-breaking):

Additionally, given some of the learnings from the PromptNode I:

• Printed the full prompt to logger.debug.
• Allow users to modify the instruction_prompt at initialization (with the default instruction prompt being the same) using a PromptTemplate. This allows for more flexibility which could be useful. Prior, we found that during internal usage we would often inject additional instructions into the query to get the output to change.

How did you test it?

Existing tests

Notes for the reviewer

Checklist

• Add additional documentation and examples in Haystack docs to help users understand how to use this node.
• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I added tests that demonstrate the correct behavior of the change
• I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test:.
• I documented my code
• I ran pre-commit hooks and fixed any issue

[tholor]

Makes all sense. I would only switch the way users can modify the prompt to be more consistent across haystack

[agnieszka-m]

I committed the changes here directly