Cannot Create Sagemaker Model for Multi-Model Enpoint. AWS docker images do not contain the required Docker Label

**Describe the bug**
This can be recreated from the AWS Console (I also confirm this occurs when calling sagemaker.model.Model.deploy()).

Using AWS Console choose Create Model (Sagemaker/Inference/Models page):

1. set "Container input options" : "Provide model artifacts and inference image location"
2. set "Provide model artifacts and inference iamge options" : "Use multiple models"
3. set "Location of inference code image" as: "763104351884.dkr.ecr.us-west-2.amazonaws.com/tensorflow-inference:2.1.0-gpu" (i also tried 2.3.1 and 2.7.0, both gave same error)
4. set "Location of model artifacts" : "I set the s3 prefix to location of my tensorflow .tar.gz models"
5. Click Create Model

The AWS Console returns this error:
ValidationException
Your Ecr Image 763104351884.dkr.ecr.us-west-2.amazonaws.com/tensorflow-inference:2.1.0-gpu does not contain required com.amazonaws.sagemaker.capabilities.multi-models=true Docker label(s).

Reproduce the failure using the AWS Console steps in the issue and the sagemaker.model.Model.deploy() path mentioned there. Check the referenced TensorFlow inference image versions and the required com.amazonaws.sagemaker.capabilities.multi-models=true Docker label. Done means establishing whether the missing label belongs in the AWS image or requires a change in this repository.

Written by the indexing model from the issue text.