feat: Enable setting a default embedding model in the stack

[franciscojavierarceo]

What does this PR do?

Enables automatic embedding model detection for vector stores and by using a default_configured boolean that can be defined in the run.yaml.

Test Plan

• Unit tests
• Integration tests
• Simple example below:

Spin up the stack:

uv run llama stack build --distro starter --image-type venv --run

Then test with OpenAI's client:

from openai import OpenAI
client = OpenAI(base_url="http://localhost:8321/v1/", api_key="none")
vs = client.vector_stores.create()

Previously you needed:

vs = client.vector_stores.create(
    extra_body={
        "embedding_model": "sentence-transformers/all-MiniLM-L6-v2",
        "embedding_dimension": 384,
    }
)

The extra_body is now unnecessary.

[ehhuang]

This should be checked when Stack was initialized instead.

[franciscojavierarceo]

I added additional validation when the stack is initialized (as well as a test to confirm) for this I think we can keep it here as well in case we allow for models to be dynamically registered and a second default were to slip in. Let me know if you'd like me to remove it though. 👍

[ehhuang]

why not just call this in _get_default_embedding_model_and_dimension?

[franciscojavierarceo]

Updated. 👍

[ehhuang]

maybe unrelated to this PR, but reading due to the change below: do we only allow one such parameter? if so, assert?

[franciscojavierarceo]

there's a break above that silently skips the others. i can add some validation to the above if we want.

[ehhuang]

non-blocking: we should formalize these parameters in a EmbeddingModel class.

[franciscojavierarceo]

yeah i can do that as a follow up

[ehhuang]

LG!