ci: run SQLite regressions on pushes and pull requests

[th3nolo]

The 12 SQLite regression tests were not running automatically, so changes to mutation protection, SQL literal handling, or result comparison could land without executing them. Add push and pull-request CI that runs the existing suite against temporary fixture databases, plus Python syntax and commit-whitespace checks.

Pin Python 3.13.7, uv 0.11.19 (release SHA-256 verified), and both actions by full commit SHA. The suite uses only the standard library; no benchmark dependencies, credentials, provider calls, or dataset downloads are needed. README documents the exact commands and fixture-only scope.

Validation:

• All 12 existing tests passed locally on Python 3.13.7 using uv 0.11.19.
• Python compileall and git diff --check passed. No formatter, linter, or type checker is configured in this repository.
• In a disposable copy, replacing the comparator with an always-true implementation caused the same unittest command to report 22 assertion failures and exit 1; repository implementation and tests were unchanged.
• Tool release identities and dates were verified against upstream GitHub metadata; all selected releases are older than 72 hours. No Python dependencies or lockfiles changed.

These checks exercise the real SQLite evaluator/comparator with small local fixtures. They do not establish measured model accuracy or benchmark-native acceptance.

GitHub validation at final head 34eee79e3ca797da0aeafc4e60212abc7f887ce8:

• Push CI: passed.
• Pull-request CI: passed; logs confirm all 12 tests executed and passed, plus syntax and whitespace checks.