Refactor OpenAI exception handling from instruction parser

[saum7800]

Description

Currently the OpenAIInstructionParser handles OpenAI exceptions from its own class. With this change, the exception is handled as required directly from the ChatGPTAgent. Refactoring the exception handling is a step in the direction of converting the OpenAIInstructionParser to a more general InstructionParser class that can be passed any Generative Agent (OpenAI, cohere, local hf models, etc).

Key Changes

• handle_openai_error returns the handled error instead of the number of API calls
• OpenAIInstructionParser checks for the response being an exception or an actual response and continues the same logic as previously specified. This pattern will be extendable to future models.
• prompt_parser test modified to mock the exact function that would be throwing the error in use.

[krrishdholakia]

oh - what was the breaking change? Happy to take a look as well @neubig

[neubig]

No worries @krrishdholakia , probably just a minor conflict: https://github.com/neulab/prompt2model/pull/328/conflicts

[saum7800]

ah yes, just a minor conflict, will update my PR

[krrishdholakia]

Ok, sounds good. Thanks for the merge!

[saum7800]

Updated the PR. Feel free to review whenever you get a chance @viswavi / @neubig