[#6505] improvement(test): Reuse core database fixtures by test class

[nevzheng]

What changes were proposed in this pull request?

Reuse the selected relational database backend for all test-template methods in a test class.

• Reset database tables before each test while retaining the class-scoped backend.
• Rebuild poisoned fixtures after setup, lifecycle, or test failures.
• Add explicit fresh-namespace isolation for tests that modify database structure.
• Preserve legacy per-invocation behavior when no backend lane is selected.
• Keep Docker containers alive for the complete test plan and surface cleanup failures.
• Avoid repeated disk cleanup when an existing MySQL or PostgreSQL container is reused.

Why are the changes needed?

Core database tests currently recreate the relational backend and schema for nearly every test-template invocation, even though the database server is already shared. This dominates lane runtime.

In the matched H2 run, this change reduced backend initialization from 460 to 48 while preserving the exact 519-test identity multiset.

Fix: #6505

Depends on: #13517

Related to: #13515

Does this PR introduce any user-facing change?

No. The fixture APIs, isolation annotations, and lifecycle changes are test-only.

How was this patch tested?

• ./gradlew :core:spotlessApply :integration-test-common:spotlessApply
• ./gradlew :core:coreUnitTest -PskipDockerTests=false
  • 1,970 passed, 2 skipped
• ./gradlew :core:coreH2Test -PskipDockerTests=false
  • 519 passed
  • 475 unique normalized identities
  • Identity digest unchanged: 22844eb5cc3bf4e6a0d85f494bb0f4a4510e2be6a3451213ac2b5f32f382f035
• python3 -B -m unittest discover -s dev/ci/tests -p 'test_core_test_identity.py'
  • 9 passed

Measurement	Before	After	Improvement
Full H2 task	324.243s	78.034s	4.15× faster
JUnit duration	188.266s	67.370s	64.2% lower
Backend initializations	460	48	89.6% fewer

MySQL and PostgreSQL timing evidence will be appended after an exclusive Docker-host validation window.