[rocksnow1942]

SpannerDialect.get_unique_constraints() read the columns of unique constraints from INFORMATION_SCHEMA.CONSTRAINT_COLUMN_USAGE, which has no column position, and returned one entry per column. A two-column constraint came back as two single-column constraints with the same name, in no particular order. On the emulator, which repeats the backing unique constraint of a foreign key once per referencing foreign key, one two-column constraint came back as eight entries.

SpannerDialect also didn't implement get_multi_unique_constraints(), so multi-table reflection (MetaData.reflect(), Inspector.get_multi_unique_constraints()) fell back to one query per table.

This PR:

• adds get_multi_unique_constraints(), which reads the unique constraints of all tables in one query with their columns from KEY_COLUMN_USAGE in ORDINAL_POSITION order, and skips the columns the emulator repeats;
• makes get_unique_constraints() call it, as get_columns(), get_pk_constraint(), get_foreign_keys() and get_indexes() do with their get_multi_*() methods;
• replaces the skipped test_get_multi_unique_constraints compliance test with a Spanner override.

A question for reviewers: Should we leave out the unique constraints?

Spanner DDL has no unique constraints, and unique indexes aren't listed in TABLE_CONSTRAINTS, so the only UNIQUE rows there are the indexes that Spanner creates to back foreign keys that refer to columns other than the primary key (IDX_<table>_<columns>_U_<hash>). get_multi_indexes() already leaves these Spanner-managed indexes out. Should the unique constraints leave them out too? Alembic's autogenerate emits op.drop_constraint() for them, which Spanner rejects with ... is not a constraint in .... This PR keeps returning them, as before.

Tested with the unit and mockserver tests and the SQLAlchemy 1.4 and 2.x compliance suites against the emulator.

• Make sure to open an issue as a bug/issue before writing your code! That way we can discuss the change, evaluate designs, and agree on the general idea
• Ensure the tests and linter pass
• Code coverage does not decrease (if any source code was changed)
• Appropriate docs were updated (if necessary)

Fixes #18836
Fixes #18837

[gemini-code-assist]

Code Review

This pull request implements the get_multi_unique_constraints method and refactors get_unique_constraints to use it, alongside adding corresponding integration and unit tests. However, a critical issue was identified in the unit tests where the test_max_size_exported function definition was accidentally deleted, causing its body to be incorrectly merged into test_get_unique_constraints.