Summary

• Make ControlConnection.wait_for_schema_agreement fall back to a same-host session pool connection when the control connection is gone, closed, defunct, busy, or raises a connection error.
• Preserve non-throwing schema agreement behavior by returning False when no same-host fallback connection is available.
• Track the last control-connection endpoint so a missing _connection can still map to the correct host pool.

Risks

• Touches control connection schema-agreement polling only.
• No protocol compatibility changes.
• If no session pool exists for the control host, schema refresh still fails softly and existing reconnect/error handling can continue.

Tests

• Added unit coverage for broken control connection fallback, gone control connection fallback, and no-fallback tolerance.
• repo-ci fast: unit phases passed; latest detailed log showed 660 passed, 77 skipped plus gevent reactor 2 passed during the cp310 wheel leg. Overall run timed out at 300s during the broader cp311 wheel leg, with no test failure reported.

Integration

• No cluster-backed integration run required for this control-flow change.