Preserve persistence ownership, legacy JSON, and literal query semantics

[sfmskywalker]

Review of the Core consolidation exposed persistence edge cases also present in Extensions main. This follow-up keeps both implementations aligned:

• Preserve legacy JsonObject documents whose fields resemble scalar or array envelopes, including polymorphic dispatch, while retaining genuine envelope compatibility.
• Make Dapper key-value saves preserve ownership: tenant-scoped value updates, primary-key-protected insertion, and narrow duplicate-key retries. A caller cannot replace another tenant's key or reassign an existing owner.
• Treat SQL prefix characters literally, including PostgreSQL backslashes and SQL Server bracket patterns.
• Compare SQLite DateTimeOffset cutoffs as instants across offsets. Sub-millisecond ties remain excluded from the exclusive liveness cutoff until a later scan.
• Apply required migration braces and update the SQLite SQL expectation.

Validation on the final source slice: 98 persistence Dapper tests and 20 generic Dapper tests passed, zero failures/skips; ownership tests use real PostgreSQL and SQLite, with a reproduced failing regression before the fix. Mongo converter tests passed 21/21 and the full Mongo suite 72/72 on the unchanged Mongo slice, including real containers. Independent source/Core code review found no blocking defects. The corresponding Core slice passed 130 persistence Dapper, 21 generic Dapper, and 89 Mongo tests.

No schema, public API, dependency, version, or publishing-workflow change. Merge with a merge commit and [skip ci] in the merge subject to avoid automatic Feedz publication.

Refs elsa-workflows/elsa-core#8623, elsa-workflows/elsa-core#8624.

[sfmskywalker]

Elsa 3 Code Review: APPROVE + HIGH @ 8626e81

Independent review by the core_integration_qa worker, reported by the integrating lead. No actionable source findings across the nine changed paths. Reviewed Mongo nominal-root envelope disambiguation and nested/genuine compatibility, literal prefix/backslash handling, SQLite instant comparison, and tenant-scoped Value-only saves with narrow duplicate-insert retries.

Independently parsed raw TRX: source persistence Dapper 98/98 and generic Dapper 20/20; mapped Core 130/130 and 21/21, zero failures/skips. The ownership matrix passed 26/26 on real SQLite/PostgreSQL, with the original regression producing three failures and one control pass. Source worktree and whitespace checks are clean. Expected Core API documentation differences were reviewed.

This is the independent code review verdict. Hosted CI was still running when this review completed and must pass separately before merge.