Add download support for modelscople.

[pass-lin]

Due to relevant legal restrictions of the Chinese government, Chinese users find it quite inconvenient to use HF.
Therefore, Chinese users often use the Chinese version of HF, which is ModelScope.
Considering this aspect, I think we should add download support for ModelScope.
Generally, Chinese users use it as a cloud disk for storing hf data. Therefore, we only need to make slight changes in the download section, and the other parts are exactly the same as hf.
The usage is similar to hf.

modelscope://username/bert_base_en

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[mattdangerw]

Thanks for the PR! We are checking this out, and also considering whether we should just build a general method to allow downloads from different sources (e.g. register a new provider).

[pass-lin]

Thanks for the PR! We are checking this out, and also considering whether we should just build a general method to allow downloads from different sources (e.g. register a new provider).

Thank you. The main reason is that we are limited by the risk of the Chinese government's control over download sources from other regions. For example, hf was initially allowed to be used in China, but later we couldn't connect to it directly, and we could only use the unstable mirror sites.
There is also a situation where Kaggle seems to have no servers in China, so the download is unstable.
Therefore, we hope to have a Chinese version available for selection, to make it more convenient for Chinese Keras users to use. We hope you can understand this.
But if your general download source still needs to connect to servers outside of China, there are still relevant risks involved.

[mattdangerw]

@pass-lin yes definitely! For sure we want there to be a way to use models with modelscope. And advantage of a "plugin" style api would be that people could also use this for third own private hosting (e.g. s3 bucket, or some custom remote file storage, etc).

Sorry the process on this one might be slightly longer since it's adding a new (soft) dependency, but we are checking this out!