fix(clickhouse): re-raise non-NOT_IMPLEMENTED errors from EXCHANGE TABLES

[wolfgang-aura]

Description

Fixes #6087.

ClickhouseEngineAdapter._exchange_tables caught every DatabaseError from EXCHANGE TABLES and only acted on NOT_IMPLEMENTED. Any other error, such as ACCESS_DENIED, was dropped. _insert_overwrite_by_condition then dropped the temp table in its finally and reported success while the target kept its old data.

The except block now re-raises unless the error contains NOT_IMPLEMENTED, as proposed in the issue. The non-atomic rename fallback moves out one indent level; its statements are the same, and test_exchange_tables covers that path.

Test Plan

Following the test plan in the issue thread, in tests/core/engine_adapter/test_clickhouse.py:

• test_exchange_tables_reraises_other_errors: execute raises an ACCESS_DENIED DatabaseError. The test asserts it propagates and that only the EXCHANGE TABLES call ran (no RENAME, no throwaway-table DROP).
• test_insert_overwrite_by_condition_replace_exchange_error_propagates: on the non-partitioned replace path, the exchange raises. The test asserts the error leaves _insert_overwrite_by_condition and the temp table is still dropped by the finally.
• test_exchange_tables (the NOT_IMPLEMENTED fallback) is not edited and passes in the same run.

Both new tests fail on unpatched main (2 failed, 34 passed) and pass with the fix (36 passed). ruff check and ruff format --check pass on both files. mypy reports nothing new on them compared with main. make fast-test was not run.

Checklist

• I have run make style and fixed any issues
• I have added tests for my changes (if applicable)
• All existing tests pass (make fast-test)
• My commits are signed off (git commit -s) per the DCO

Written with Claude (claude-sonnet-5-5) and reviewed by a second Claude instance under the Mailman harness.