🐛 Bug: NoSuchTableError is raised if user tries to access a stream that did not receive records

[aaronsteers]

Currently we only run 'create table' when and if record batches are being processed for that stream. Since we know which stream names are selected, and since we have the schema declared in the catalog for those streams, we could optionally also create empty destination tables and initialize empty CachedDatasets for those streams.

This would be helpful because a user does not expect to get an error when checking for the results of a specific stream. Example, the "deployments" stream of the GitHub source is very likely to be empty. A user iterating through the streams they selected will get a confusing failure rather than just seeing an empty dataset.

Example error from Github (internal slack thread here):

sqlalchemy.exc.NoSuchTableError: deployments