Fix bulk shift ordering to prevent duplicate key violations on MySQL

[jasper-vandemalle]

Fixes #464.

When shifting list indices upward (inserting before the end of a list), MySQL processes the bulk UPDATE ... SET idx = idx + 1 in ascending primary key order. The row at the lowest index gets updated first, and its new value collides with the next row's current index.

This adds ORDER BY idx DESC to the bulk shift statement for positive shifts, so the highest-index row moves first and each update lands in a vacant slot. Negative shifts (from remove operations) already work correctly with ascending order.

ORDER BY in UPDATE is a MySQL/MariaDB extension not supported by PostgreSQL, Oracle, or SQL Server, so this is gated behind a new ORDER_BY_IN_UPDATE_STATEMENT adapter capability. MySQL and H2 register it; other databases keep the existing behavior.

Also fixes a minor bug in the internalShiftBulk catch block that referenced shiftStmt instead of the local shiftBulkStmt.

Also included an additional integration test:

• Add regression test for join-table list bulk shift ordering tests#91