Reasoning Models

I'll work on this during Drupalcon ATL sprint.

So here's what I did:

• Simple notification in the Default Providers: I added a notification for all providers using Groq to inform them about the available settings to override defaults.
• Expanded Groq Configuration Form: The Groq config form now includes defaults such as temperature, max tokens, reasoning, and JSON output options. Also we offer an override-per-operation option - to override these defaults (see below).
• Operation-Specific Overrides: There are two operations - chat and chat_with_json - where users can globally override settings. Fallback settings have also been implemented.
• Enhanced Groq Model Usage: I expanded the use of the Groq model(s). Chat with JSON is now designed so that any model can be used (it's the dev-user that should decide). I also removed models related to speech, keeping only the textual models for now (additional updates will follow in future tickets).
• Configuration Persistence: I ensured that the configurations saved are indeed applied during runtime.

Testing Process

1. Enable the module, configure it, and set up the necessary API keys.
2. Enable the assistant and chatbot modules.
3. Use the chatbot with an assistant:
   • Select a reasoning model in the default settings (or assistant settings) to verify that the model's thought process is displayed.
   • Return to the Groq module settings, change the reasoning option to a hidden value, refresh the interface, and confirm that only the final results are shown - omitting the internal reasoning.

+ I also added extra settings in the api_defaults.yml - to make sure that assistants can override these (we do an array_merge to make sure it's good).

NOTE: I did this against ai.module released version - NOT the DEV (I will need to make sure it aligns with DEV later).

Changing to 1.1.x-dev - we could also backport it and create a first 1.0.0-rc1 if wanted?

Thanks, getting merged

Automatically closed - issue fixed for 2 weeks with no activity.