Fix TableEngine construction for alembic-rendered zero-arg engines and SummingMergeTree columns

[agu2347]

Closes #946.

Two independent bugs in clickhouse_connect/cc_sqlalchemy/ddl/tableengine.py:

1. Zero-arg engines can't be constructed from their own repr().

Memory, Log, StripeLog, TinyLog, Null, and Set don't define their own __init__, so they fall through to TableEngine.__init__(self, kwargs), which had no default for kwargs. repr(Memory({})) already produced "Memory()", but Alembic's autogeneration writes exactly that no-arg call into the migration file, and running that migration executes Memory() -- which raised TypeError: __init__() missing 1 required positional argument: 'kwargs'.

Fixed by giving kwargs a None default, normalized to {} inside the constructor.

2. SummingMergeTree has no way to accept a column list.

ClickHouse's SummingMergeTree([columns]) engine takes an optional list of columns to sum on merge (e.g. SummingMergeTree((delta, n_tx)) ORDER BY id), but the Python class was a bare class SummingMergeTree(MergeTree): pass -- no columns parameter existed anywhere in the constructor chain.

Fixed by giving SummingMergeTree its own constructor, following the same pattern already used by ReplacingMergeTree/CollapsingMergeTree/etc: extend TableEngine directly (not MergeTree, whose narrower signature has no room for columns), add columns to arg_names as an optional positional arg, and extend TableEngine.__init__'s positional-arg rendering to accept a tuple/list value and render it as a parenthesized column list -- mirroring how eng_params already renders tuples for ORDER BY/PARTITION BY.

Testing

Added 6 tests to tests/unit_tests/test_sqlalchemy/test_alembic.py:

• test_no_arg_table_engines_accept_zero_args -- all six zero-arg engines construct and round-trip.
• test_memory_alembic_repr_round_trips -- the exact scenario from the issue: repr(Memory({})) produces code that evals back cleanly.
• test_summing_merge_tree_accepts_columns -- SummingMergeTree(columns=(...), order_by=...) compiles to valid DDL.
• test_summing_merge_tree_columns_optional -- columns remains optional.
• test_summing_merge_tree_requires_order_by_or_primary_key -- existing MergeTree-family constraint preserved.
• test_summing_merge_tree_repr_round_trips -- repr -> eval round trip with columns.

All 455 unit tests in tests/unit_tests/test_sqlalchemy/ pass (449 existing + 6 new), confirmed with no regressions. Reverting the fix makes 5 of the 6 new tests fail with the exact symptoms described in the issue (the 6th doesn't exercise the changed code path, so it correctly still passes). ruff check is clean on both changed files.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ joe-clickhouse
❌ agu2347
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[joe-clickhouse]

Hi @agu2347 looks good! I added a few finishing touches that do the following:

• Preserves the existing SummingMergeTree positional API and MergeTree inheritance
• adds matching columns= support for ReplicatedSummingMergeTree
• validates column-list inputs
• address missing type declarations
• adds the CHANGELOG entry

To get this merged, please sign the CLA. thanks!