[ddl] Distributed ADD INDEX hangs in running/retry on persistent SetTSBeforeImportEngine engine-not-found errors

## Bug Report

> Evidence status: confirmed.

### 1. Minimal reproduce step (Required)

On testbed 8220955 with tidb_enable_dist_task=ON and tidb_ddl_enable_fast_reorg=ON, use a commit-matched failpoint owner and inject github.com/pingcap/tidb/pkg/ingestor/ingestctrl/mockAINativeSetTSBeforeImportEngineErr=return("engine_not_found") before SetTSBeforeImportEngine on a 5k-row ADD INDEX. Baseline job 2313 synced. One-shot control job 2322 synced. Persistent fault keeps job 2319 / global task 270003 in running/write reorganization until the failpoint is removed.

### 2. What did you expect to see? (Required)

A source-native fundamental import/setup error such as engine-not-found should fail or cleanly roll back the distributed subtask instead of being retried forever.

### 3. What did you see instead? (Required)

Lightning/ingest side logs the error as un-retryable, but the DXF task executor treats the subtask as retryable/idempotent and reruns it, leaving the DDL job running until external fault removal.

### 4. What is your TiDB version? (Required)

TiDB v9.0.0-beta.2.pre-1895-g5c9198e948 distributed ADD INDEX / DXF ingest path

Likely root cause and fix direction

backfillDistExecutor.IsRetryableError falls through to isRetryableError(err, true), so unknown/source-native runtime errors from the import path are treated as retryable. task_executor keeps the subtask in running state and reruns it forever.