fix: Unregister a model from registry if not being served

[terrytangyuan]

What does this PR do?

This ensures that the model is deleted from the registry if it's not actively being served by vLLM remote server.

Test Plan

All tests in tests/client-sdk/inference/test_text_inference.py passed.

LLAMA_STACK_BASE_URL=http://localhost:5002 pytest -v tests/client-sdk/inference/test_text_inference.py
=============================================================== test session starts ===============================================================
platform linux -- Python 3.10.16, pytest-8.3.4, pluggy-1.5.0 -- /home/yutang/.conda/envs/distribution-myenv/bin/python3.10
cachedir: .pytest_cache
rootdir: /home/yutang/repos/llama-stack
configfile: pyproject.toml
plugins: anyio-4.8.0
collected 16 items                                                                                                                                

tests/client-sdk/inference/test_text_inference.py::test_text_completion_non_streaming[meta-llama/Llama-3.1-8B-Instruct] PASSED              [  6%]
tests/client-sdk/inference/test_text_inference.py::test_text_completion_streaming[meta-llama/Llama-3.1-8B-Instruct] PASSED                  [ 12%]
tests/client-sdk/inference/test_text_inference.py::test_completion_log_probs_non_streaming[meta-llama/Llama-3.1-8B-Instruct] PASSED         [ 18%]
tests/client-sdk/inference/test_text_inference.py::test_completion_log_probs_streaming[meta-llama/Llama-3.1-8B-Instruct] PASSED             [ 25%]
tests/client-sdk/inference/test_text_inference.py::test_text_completion_structured_output[meta-llama/Llama-3.1-8B-Instruct] PASSED          [ 31%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_non_streaming[meta-llama/Llama-3.1-8B-Instruct-Which planet do humans live on?-Earth] PASSED [ 37%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_non_streaming[meta-llama/Llama-3.1-8B-Instruct-Which planet has rings around it with a name starting with letter S?-Saturn] PASSED [ 43%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_streaming[meta-llama/Llama-3.1-8B-Instruct-What's the name of the Sun in latin?-Sol] PASSED [ 50%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_streaming[meta-llama/Llama-3.1-8B-Instruct-What is the name of the US captial?-Washington] PASSED [ 56%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_with_tool_calling_and_non_streaming[meta-llama/Llama-3.1-8B-Instruct] PASSED [ 62%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_with_tool_calling_and_streaming[meta-llama/Llama-3.1-8B-Instruct] PASSED [ 68%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_with_tool_choice_required[meta-llama/Llama-3.1-8B-Instruct] PASSED [ 75%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_with_tool_choice_none[meta-llama/Llama-3.1-8B-Instruct] PASSED [ 81%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_structured_output[meta-llama/Llama-3.1-8B-Instruct] PASSED     [ 87%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_tool_calling_tools_not_in_request[meta-llama/Llama-3.1-8B-Instruct-True] PASSED [ 93%]
tests/client-sdk/inference/test_text_inference.py::test_text_chat_completion_tool_calling_tools_not_in_request[meta-llama/Llama-3.1-8B-Instruct-False] PASSED [100%]

==================================================== 16 passed, 1 warning in 586.72s (0:09:46) ====================================================

[terrytangyuan]

@ashwinb I saw a need to do this in Ollama as well (noticed your commit 36162c8) since we want to undo the model registration if it's not available. WDYT?

[ashwinb]

@terrytangyuan yeah I think we need to make this work better -- we should not put the model into the registry at all until the provider acknowledges successfully that they have registered it. I am going to refactor ModelRegistryHelper slightly (it has become a convoluted mess (with so little code) with badly named and behaved methods)

[terrytangyuan]

@ashwinb I just applied the same to Ollama as well. I think this will at least correct some existing behavior before your refactoring.