[Cosmos] Service Request and Response retries

[simorenoh]

This PR introduces cross-regional retries for requests failing to reach the service, or where the service failed to provide a response.

Changes
In the case of a ServiceRequestError, where we know the request didn't reach the service, we know it is safe to retry. In this case we don't have any filters for the operation, and retry once in each relevant preferred region location before raising the exception to the user.

In the case of a ServiceResponseError, where the service was not able to give a response, we filter on these being only exceptions that have to do with unavailability due to timeouts with the service. If that's the case, then we filter on these requests being exclusively document-level read operations, since these are the only ones we know we can retry safely upon not being able to ascertain the output of the request sent. These are retried in the relevant available preferred read locations.

This PR also contains a bugfix to ensure that even if preferred regions are not present in the current account regions, we don't attempt to go into them.

Testing
Manual testing for the first set of changes has been done by reducing the connection timeout on requests after the client is initialized and fetches the database account information. Added tests mocking this functionality.

Manual testing for the bugfix has been done as well, where we see that even though the most_preferred_endpoint is one that is no longer present in the account, we don't attempt to go into it when resolving the service endpoint.

[azure-sdk]

API change check

APIView has identified API level changes in this PR and created following API reviews.

azure-cosmos

[jeet1995]

LGTM - thanks!