[Cosmos] move is_system_key logic to be computed only once on container usage

[simorenoh]

A user reported the fact that the async client couldn't properly deal with single partition queries since the result of the call to is_system_key was not being awaited properly. Due to this, we have a mitigation, but it is not ideal: #22893

We are currently awaiting the result of this network call once for every query being executed with a partition key, adding overhead for each query when it is only needed once in order to populate the container settings and then be used from there. Have tried a couple different things but haven't gotten it to work yet.

Expected outcome would be to run this network call only once as stated before - need to find a way to do so.

User issue in question: #22861