Support Sentence Transformer Models for text (non embedding)

Is your feature request related to a problem? Please describe.
Currently we have custom code to do mean pooling / last token. However that doesn't work for models like google/embeddinggemma-300m which do linear projection before doing mean pooling (and the linear projection is not available on AutoModel).
The code will be simpler if in our load_model we do SentenceTransformers.load_model(...)

I like how vLLM does it in their test cases which is is_sentence_transformer=??? to correctly load either from AutoModel or SentenceTransformer (https://github.com/vllm-project/vllm/blob/11857a00b0a59183286eab393df4e13b20efec3a/tests/conftest.py#L279)

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.