AAP-51301: Modify ansible-chatbot-stack build to include dependencies for gemini provider

[romartin]

Jira Issue: https://issues.redhat.com/browse/AAP-51301

Description

Setting up the Gemini inference provider into the stack. It will be used by the Ansible AI Assisted Installer, the ALIA one will keep using Granite.

Just notice that from this point, Gemini models are now being listed in the v1/models endpoint as well (but only working if an API key is properly set):

        {
            "identifier": "gemini/gemini-1.5-flash",
            "metadata": {},
            "api_model_type": "llm",
            "provider_id": "gemini",
            "type": "model",
            "provider_resource_id": "gemini-1.5-flash",
            "model_type": "llm"
        },
        {
            "identifier": "gemini/gemini-1.5-pro",
            "metadata": {},
            "api_model_type": "llm",
            "provider_id": "gemini",
            "type": "model",
            "provider_resource_id": "gemini-1.5-pro",
            "model_type": "llm"
        },
        {
            "identifier": "gemini/gemini-2.0-flash",
            "metadata": {},
            "api_model_type": "llm",
            "provider_id": "gemini",
            "type": "model",
            "provider_resource_id": "gemini-2.0-flash",
            "model_type": "llm"
        },
        {
            "identifier": "gemini/gemini-2.5-flash",
            "metadata": {},
            "api_model_type": "llm",
            "provider_id": "gemini",
            "type": "model",
            "provider_resource_id": "gemini-2.5-flash",
            "model_type": "llm"
        },
        {
            "identifier": "gemini/gemini-2.5-pro",
            "metadata": {},
            "api_model_type": "llm",
            "provider_id": "gemini",
            "type": "model",
            "provider_resource_id": "gemini-2.5-pro",
            "model_type": "llm"
        },
        {
            "identifier": "gemini/text-embedding-004",
            "metadata": {
                "embedding_dimension": 768.0,
                "context_length": 2048.0
            },
            "api_model_type": "llm",
            "provider_id": "gemini",
            "type": "model",
            "provider_resource_id": "text-embedding-004",
            "model_type": "llm"
        },

Testing

Tested locally.

Production deployment

• This code change is ready for production on its own
• This code change requires the following considerations before going to production:

[TamiTakamiya]

I have tested with the container image built in the Konflux PR build. Gemini (gemini-2.5-light) LLM was used properly, but for some reasons the knowlege_search failed. It may be caused by some setting issues of my test environment...

[TamiTakamiya]

I have tested with the container image built in the Konflux PR build. Gemini (gemini-2.5-light) LLM was used properly, but for some reasons the knowlege_search failed. It may be caused by some setting issues of my test environment...

It was a configuration issue on my test environment set-up. I could call /streaming_query endpoint and get referenced documents list both with Granite 3.3 and gemini-2.5-flash using the container image quay.io/ansible/ansible-chatbot-stack:on-pr-86-22765790b0b63c98cf25042309d3a08033654fa5

[TamiTakamiya]

Although I left a comment on uv export format, it's very minor. I have verified the container image built worked with Gemini (and Granite) LLM.

After the conflict is resolved, I will re-approve this. Thanks!

[romartin]

@ldjebran @TamiTakamiya all changes from your comments applied, also rebased from main Ready for final review plz!

[ldjebran]

LGTM

[TamiTakamiya]

LGTM