op-acceptance-tests: run TestNotTruncateDatabaseOnRestartWithExistingDatabase against kona-node

## Summary

[`TestNotTruncateDatabaseOnRestartWithExistingDatabase`](https://github.com/ethereum-optimism/optimism/blob/develop/op-acceptance-tests/tests/safeheaddb_elsync/safeheaddb_test.go) is skipped when the acceptance suite runs against kona-node.

After restarting the CL, the test cannot read the persisted safe-head data it expects from kona-node.

## Required work

Reproduce the restart path, determine whether the gap is Kona persistence, RPC parity, or test synchronization, fix the responsible layer, and remove the Kona skip.

## Acceptance criteria

- [ ] The root cause and resolution are documented.
- [ ] `TestNotTruncateDatabaseOnRestartWithExistingDatabase` no longer skips when `DEVSTACK_L2CL_KIND=kona-node`.
- [ ] The test verifies that existing safe-head data survives a kona-node restart.
- [ ] Existing op-node coverage continues to pass.

## Related

- https://github.com/ethereum-optimism/optimism/issues/20278
- https://github.com/ethereum-optimism/optimism/issues/20649
- Parent inventory: https://github.com/ethereum-optimism/protocol-team/issues/101