[BUG] Deprecated models are used for error handling during failed tool calls in Next environment

[SamvelSimonyan]

Steps to reproduce:

1. Run an agent or pipeline in the Next environment.
2. Make a tool call that fails.

Actual results:

• When the tool call fails, low-tier LLM and sometimes high-tier LLM are called to debug the tool call as part of error handling.
• Deprecated models like GPT-4o are used, causing issues as they do not work.

Expected results:

• Elitea should automatically choose the default low-tier model from the configuration for error handling.
• If the default model is not selected or is unavailable, Elitea should select the next available model from the list, ensuring that tuned-off or deprecated models are not used.

Other info:

• The issue is specific to error handling during failed tool calls in the Next environment.
• Proper model selection logic should be implemented to avoid using deprecated models.

[SamvelSimonyan]

e.g., in the following apipelien Gpt 5 is used for troubleshooting in some stage, but it is removed from the models list ( not completely removed but turned off )

https://next.elitea.ai/app/27/pipelines/all/334?viewMode=owner&name=Advanced%20BA%20Assistant

[SamvelSimonyan]

A simple solution will be to delete the models completely from elitea m, but the bug will remain