[FEATURE]Release Asynmetric Model from Opensearch Model Server

[dhrubo-os]

Is your feature request related to a problem?

Currently ML-Commons support Asymmetric embedding for models. Example Blue print PR

But in the blue print, if you notice, we need to use another hugging face url to get the onnx format of the model and the user need to zip the file and register it to the ml-commons which is a burden for customer.

Like other pre-trained embedding models, we can release another Asynmetric model from our model release workflow for customers.

What solution would you like?
A clear and concise description of what you want to happen.

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

Do you have any additional context?
Add any other context or screenshots about the feature request here.