Raise exceptions

[atesgoral]

Note
This PR builds on #338. I unfortunately can't set the base when cutting a PR from a forked repo. Reviewing it commit by commit after skipping the first commit would be helpful.

• Proper SSE parsing #338

• BREAKING CHANGE: ruby-openai now raises errors instead of silently ignoring them. It is better for production apps to fail fast and take note of any errors instead of ignoring them. OpenAI is not expected to return any invalid payload in its normal operation.
• Removed the completion test case from the finetunes suite because it wasn't adding any value since it doesn't really exercise an actual fine-tuned model. Besides, the create cassette was broken (had 400 responses throughout). The reason that test was passing is because ruby-openai was eating those errors.
• Tweaks to README:
  • Tweak/add some bits about token counting during streaming
  • Replace "ChatGPT" with "GPT" (the former is a web app, the latter is the actual model)

All Submissions:

• Have you followed the guidelines in our Contributing document?
• Have you checked to ensure there aren't other open Pull Requests for the same update/change?
• Have you added an explanation of what your changes do and why you'd like us to include them?

[alexrudall]

Thanks @atesgoral - taking a look now

[alexrudall]

Thanks a lot for this!

[alexrudall]

Fixed a couple of errors but that spec is still failing due to Faraday raising the error 🤔

[atesgoral]

@alexrudall We might need to pass in a full-fledged mock Request object. I'll poke around.

[alexrudall]

@alexrudall We might need to pass in a full-fledged mock Request object. I'll poke around.

Legend, thank you

[atesgoral]

Ah, I just remembered removing that HTTP error case from the http spec because I was running into issues with mocking a request. I then added a VCR at the client spec level instead.

The old test might have popped back in due to a bad rebase. I removed it again.