Fix: embedding serving setup error in LiteLLMServing

[Libao-Yang]

LiteLLMServing supports embedding services and implements real embedding calls through the generate_embedding_from_input function. However, an error occurs during the initialization process. The error log is as follows:

During handling of the above exception, another exception occurred:
.venv/lib/python3.11/site-packages/litellm/main.py:3773: in completion
    response = base_llm_http_handler.completion(
.venv/lib/python3.11/site-packages/litellm/llms/custom_httpx/llm_http_handler.py:524: in completion
    response = self._make_common_sync_call(
.venv/lib/python3.11/site-packages/litellm/llms/custom_httpx/llm_http_handler.py:246: in _make_common_sync_call
    raise self._handle_error(e=e, provider_config=provider_config)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.venv/lib/python3.11/site-packages/litellm/llms/custom_httpx/llm_http_handler.py:4507: in _handle_error
    raise provider_config.get_error_class(
E   litellm.llms.ollama.common_utils.OllamaError: {"error":"\"all-minilm:latest\" does not support generate"}

The root cause is that LiteLLMServing.__init__() forcibly executes _validate_setup(), and _validate_setup() always calls litellm.completion(), which leads to a setup error when using an embedding model.

[SunnyHaze]

LGTM, Thanks!