Fix JSON field match

[mmabrouk]

Closes: AGE-1537
Closes: #2497

Description

The issue was that the setting values were not 1) propagated to the evaluation function when the call came from the backend 2) the json_field was not used to extract the field from the json. This PR fixes both issues

QA:

The following has been tested:

• Regression: removing the settings from the input and making sure the new sanity checks work and propagate to the backend
• The evaluator work when run from the backend in case of True (field exists and it is the same as correct_asnwer)
• Works, in case the llm output is not json
• Works, in case the llm output does not contain the field
• Works in case the llm output contains the field but the result is not different

What is still to be QA'ed:

• Does the interface work from the UI test interface in cloud

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
agenta	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Feb 20, 2025 11:05am
agenta-documentation	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Feb 20, 2025 11:05am

[jp-agenta]

lgtm!