Follow-up to #3582652, which fixed a defect that did not exist and introduced a real one. ## What #3582652 assumed That the ECA action or condition item surrounding `configuration` never carries a populated `original_id`, so the schema expression `tamper.[%parent.original_id]` could never resolve. ## What is actually the case ECA populates it. `Drupal\eca\Entity\Eca::addAction()` reads `original_id` from the plugin definition and writes it onto the item: ```php $original_id = $this->actionPluginManager()->getDefinition($plugin_id)['original_id'] ?? NULL; if ($original_id !== NULL) { $this->actions[$id]['original_id'] = $original_id; } ``` `addCondition()` does the same, and `eca.schema.yml` declares `original_id` on both item mappings. Both arrived in ECA with #3541352, "Allow plugin derivers to derive config schema as well" — this exact mechanism, already provided. Since `configuration` is typed `action.configuration.[%parent.plugin]`, `%parent` is the item, and the item has `original_id`. The original expression was correct. ## The consequence #3582652 changed the two sections to `tamper.[original_id]` and made `TamperTrait` write the key into the stored plugin configuration. On a site running 2.0.x dev, saving a model now stores `original_id` twice: once on the item, where ECA puts it, and once inside `configuration`. The second one is redundant and is only there to satisfy a schema change that was not needed. No tagged release carries this, so only sites tracking the branch are affected. ## The real defect behind the original report Models whose configuration was authored as YAML rather than saved through ECA's entity API never pass through `addAction()`, so they have no item-level `original_id` and the schema legitimately cannot resolve the wrapped tamper plugin. The same applies to models stored before #3541352. That is a data gap, not a schema gap, and the fix is to put the key on the item. ## Proposed resolution 1. Revert the schema sections to `tamper.[%parent.original_id]` and drop the `original_id` declaration from both mappings. 2. Revert the `TamperTrait` changes: the write in `tamperDefaultConfiguration()` and both strip sites. 3. Re-point `eca_tamper_post_update_add_original_id()` so it removes the stray `configuration.original_id` from models stored by branch code, rather than adding it. 4. Correct `ConfigSchemaTest` to build its fixture the way ECA actually stores a model, with `original_id` on the item. With that fixture the released schema passes unaided, which is the evidence the original issue lacked. Everything else from #3582652 stands and is unaffected: the `#[LegacyHook]` wrapper that finally runs the hook class on Drupal 10.4, the relocation of the kernel test out of `src/`, the ECA 2 / ECA 3 kernel test base, and `OPT_IN_TEST_PREVIOUS_MAJOR`. A generic backfill of the item-level key for models that predate #3541352 belongs in ECA rather than here, and is filed separately.