Allow Model.upload not to use any serving contianer

Currently `aiplatform.Model.upload()` requires user to specify the `server_container_image_uri` ([link](https://cloud.google.com/vertex-ai/docs/samples/aiplatform-upload-model-sample#aiplatform_upload_model_sample-python)):
```
model = aiplatform.Model.upload(
display_name=DISPLAY_NAME,
artifact_uri=URI,
serving_container_image_uri="gcr.io/cloud-aiplatform/prediction/pytorch-cpu.1-9:latest"
)
```
It is clear the requirement comes from AutoML platform to allow model for ease of Endpoint deployment (which is a great and nicely streamlined feature!). However, it would be really optimal to allow flexibility of no needing to specify the serving container.

Particular need comes when user just wants to use VertexAI as Tracking and Registry service without fully leveraging AutoML with further Endpoint deployment (which requires embedded HTTP server within the service container).

It would be really convenient to the user to just be able to register the model for further consumption without having to bother about service capability. Current workaround is to still specify a dummy service container as above, but not having to assign any container resources would be better.

Thanks!

Start at the aiplatform.Model.upload entry point and trace how serving_container_image_uri is validated and passed to Vertex AI. Clarify the registration-only behavior and its API constraints, then add coverage showing that a model can be registered without a serving container while preserving the existing deployment path.

Written by the indexing model from the issue text.