[dajiaohuang]

Description

Keep the catalog selected by [catalog.]database consistent across connections:

• Starting with athenacli custom.analytics stores analytics as the database, but the completion refresher creates a new executor with only that database. It therefore reconnects to AwsDataCatalog, so completion metadata comes from a different catalog than queries.
• use other.reporting connects to other but does not update the stored catalog. A subsequent use archive, reconnect, or completion refresh returns to the old catalog.

Preserve the catalog after a successful database change and pass it explicitly into the completion executor. The default remains AwsDataCatalog; qualified database arguments still select the catalog. This follows the catalog/USE behavior introduced by #82.

Validation: offline pytest on Python 3.11: 58 passed, 1 XPASS, including four new tests for refresh, database changes/reconnect, the default catalog, and failed connection state. Also exercised the public CLI and AthenaCli.change_db with an owned PyAthena connection mock; baseline refresh selected AwsDataCatalog, corrected refresh selected custom.

All execution used isolated HOME/config, synthetic credentials, disabled SDK credential discovery and blocked network connections. No real Athena/S3/IAM request or paid query was made; live AWS and Python 3.6–3.8 acceptance are untested. Existing deprecation warnings remain.

Checklist

• I've added this contribution to the changelog.md.
• I've added my name to the AUTHORS file (or it's already there).