Base model management in runtime for Orchestrator

[boydc2014]

Related to #4209.

Given the size of base model is pretty large (400+ MB for one language), it makes sense to let Composer app and runtime both manage it's own model downloading, saving, refering without have to exchange base model(s).

This ticket to track the work on runtime side, including

• pick location to store base models
• detect the base model version from bot assets
• auto-download the base model when version detected
• auto-update the model path to point to the location picked in step 1

The runtime effort needs to consider a few others scenarios

1. the impact to azure deploying.
2. the impact to ejecting\exporting runtime.

[boydc2014]

After checking with Tai and Leilei, we are closing this because we already implemented a solution that avoid unnecessary copy in local environment. For azure publishing, the best solution we are visioning, using a shared drived mounted in different web app instances, is not achievable at the moment, because Azure Blob\Files do not support Windows web app. No action needed at this moment.