[AGE-272] [Bug] Improve errors in evaluation view

[mmabrouk]

When errors happen in the evaluation view, there are a couple of problems:

• The error in the llm app are shown in black and not in red as expected
  • The issue is in the LLM app response and not in the UI
• The error in the llm app is very general (500 internal error) and does not provide details about the message in the error as we do have in the playground (❌ Model gpt-3.5-turbo does not support JSON response format)
  • Unclear whether this is still relevant
• The error for the evaluation is the same as in the llm app. Instead we should say "LLM app failed" or something similar.
  • This is resolved by the open PR

_{From SyncLinear.com | AGE-272}

[mmabrouk]

@jp this issue should actually be fixed in the new version of the SDK (we still are returning things as 200 messages but with another output). However the templates have not been updated / deployed yet.

[mmabrouk]

not sure tbh on the interface here

[mmabrouk]

My guess is that FastAPI should return 401 with details