[Zxf-xufeng]

Which issue or RFC does this PR close?

Closes #2160.

Rationale for this change

Schema reflection rejected PowerContext's own pc_recall_effort_daily CHECK (dropped_items = dropped_below_min_bytes + dropped_no_fitting_truncation) as having unbalanced parentheses. The lexer recognized addition, but the expression parser stopped before it, preventing migration status from reaching layout identification.

What changes are included in this PR?

• Parse binary +, -, *, /, and % with arithmetic precedence above comparisons and boolean operations. Preserve left associativity, unary signs, and nested grouping when rendering validation SQL.
• Parse balanced subexpressions independently. Unsupported CHECK syntax reports table, named constraint, and construct context without echoing string literals.
• Execute parameter-free CHECK validation SQL with statement-scoped no_parameters=True, so the MySQL driver preserves remainder operators and percent signs inside literals. Constraint violations still raise data_integrity_error.
• Add actual seekDB SHOW CREATE TABLE fixtures for three tables recreated from the 1.3.0 tag's metadata and limits, recording source blob IDs, compiler/binding versions, and engine version.
• Cover expression semantics, malformed/unsupported syntax, historical reflection, and real seekDB execution. Real-driver regressions exercise both MySQLSchema.validate_data() and MySQLMigrationRunner.plan() for remainder and percent-literal CHECKs, including valid, NULL, and invalid stored values.
• Include the CHECK test module in the existing real seekDB CI job for Python 3.11 and 3.14. Document parser behavior and fixture provenance.

Are there any user-facing changes?

Migration schema inspection can parse the reported arithmetic CHECK and continue to the actual baseline/support decision. Unsupported CHECK expressions report useful schema context; remainder validation no longer fails in driver parameter interpolation.

There are no API, persisted-schema, dependency, or migration-baseline changes. No tables or constraints are excluded. Successful reflection does not enable upgrades of unregistered complete Server layouts or bypass revision verification, recovery eligibility, and data-integrity checks.

How was this change tested?

• Reproduced TypeError: not enough arguments for format string on the original PR head through the real seekDB aiomysql bridge for both a remainder CHECK and a percent-containing string literal.
• POWERCONTEXT_TEST_MIGRATION_SEEKDB=1 .venv/bin/python -m pytest -q tests/builtin/persistence/test_mysql_migration_checks.py tests/builtin/persistence/test_mysql_migration_schema.py --tb=short — 58 passed.
• .venv/bin/python -m pytest -q tests/builtin/persistence/test_*migration*.py tests/test_database_migration*.py tests/test_processing_migration_cli.py --tb=short — 164 passed, 33 skipped (opt-in backend tests).
• UV_NO_SYNC=1 UV_PYTHON=3.14 make check — passed, including lock consistency, all pre-commit hooks, type checks, workflow/integration consistency, and 50 integration-manifest tests.
• Confirmed that the CI -k seekdb selector includes both new driver regressions and the historical CHECK tests.

Local validation used macOS, Python 3.14.6, SQLAlchemy 2.0.51, aiomysql 0.3.2, and pylibseekdb 1.4.0.post1. The historical fixture was captured on seekDB v1.4.0.0 using 1.3.0 definitions; it is not a customer database or a reproduction of a complete customer upgrade. No live OceanBase test was run. CI results must be checked separately.

AI usage statement

OpenAI Codex inspected the issue and migration contracts, implemented the parser and driver fixes, added regression coverage and CI selection, and ran local validation and a full diff review.

[hidb4ai]

The reported addition case is fixed. One issue remains in the newly added remainder support; details inline.

[hidb4ai]

LGTM