fix(sessions): use LONGTEXT for MySQL message storage

[rioyu123]

Summary

This pull request changes the message_data column to LONGTEXT for newly created MySQL and MariaDB message tables. Other dialects retain TEXT; serialization, session identifiers, transactions, and public APIs are unchanged. Existing tables are not migrated.

MySQL-family TEXT storage is too small for some serialized session items. In a live example, Chinese/emoji content serialized to 1,200,028 bytes: strict mode rejected the TEXT insert with error 1406, while non-strict mode logged truncation and stored 65,535 bytes of invalid JSON that the SDK could not return as the original item. A LONGTEXT column preserved the full item through SDK add/get/pop in both modes.

Dependency status: #4741 was closed without merging because MySQL/MariaDB schema-creation work is not a current maintainer priority. This draft is being closed as well. It contains none of #4741's changes and does not independently enable automatic table creation. Existing caller-managed schemas remain unchanged.

Any future continuation would need an accepted schema-creation baseline, fresh MySQL/MariaDB automatic-table-creation and small/large-item tests, and the full verification and review steps. The results below cover the draft as tested, not that combined path.

Test plan

• tests/extensions/memory/test_sqlalchemy_session.py: 47 passed, including four new column-compilation cases for MySQL, MariaDB, PostgreSQL, and SQLite.
• Ruff check and format check passed for both changed files; git diff --check passed.
• Real MySQL 8.0.46 and MariaDB 11.8.9, using caller-managed tables and SQLAlchemySession(create_tables=False): small-item controls passed; strict TEXT raised error 1406; non-strict TEXT produced truncated, invalid JSON; LONGTEXT preserved the complete 1,200,028-byte serialized value and exact SDK get/pop results in both modes.
• Offline complete-table compilation: current main plus this change still fails on unbounded session_id; fix(sessions): support MySQL and MariaDB schema creation #4741's schema plus the proposed column type compiles for both MySQL and MariaDB. This was a disposable metadata composition check, not a combined-source live automatic-creation test.
• The full repository verification stack and final schema-change review were not completed for this draft. No full-stack pass is claimed.

Issue number

N/A. Related schema-creation work: #4741.

Checks

• I've added new tests, if relevant
• I've run .agents/skills/code-change-verification/scripts/run.sh
• I've confirmed all verification steps pass
• If using Codex, I've run /review before submitting this PR

[github-actions]

This PR is stale because it has been open for 10 days with no activity.

[rioyu123]

Still planned, and intentionally kept as a draft while #4741 (or an equivalent MySQL/MariaDB schema-creation fix) is reviewed. That dependency has not merged yet. Once the accepted fix reaches main, I will update this branch and verify automatic table creation plus small- and large-item round trips on both databases before requesting review.

Could a maintainer add skip-stale while that dependency is pending? I would like to keep the column-capacity fix separate from the schema-creation work.

[rioyu123]

Closing this draft following the decision on #4741. The LONGTEXT change depends on an accepted MySQL/MariaDB schema-creation baseline, so it cannot move forward on its own. The patch and test results remain available here if this work is revisited. Thanks for the consideration.