ES|QL's streaming Apache Arrow helper needs better handling of server-side errors

### 🐛 Bug report Reported by @cnasikas: > We’re using the JS `esClient.helpers.esql().toArrowReader()` helper to stream Arrow IPC results. When the ES|QL query produces errors like unsupported field types like `date_range`, Elasticsearch correctly returns a JSON error ``` (content-type: application/vnd.elasticsearch+json;compatible-with=9) {"error":{"root_cause":[{"type":"illegal_argument_exception","reason":"ES|QL type [date_range] is not supported by the Arrow format"}],"type":"illegal_argument_exception","reason":"ES|QL type [date_range] is not supported by the Arrow format"},"status":400} ``` but the helper still passes this response into `AsyncRecordBatchStreamReader`, which then throws a misleading Arrow metadata error `("Expected to read … metadata bytes")`. ### To reproduce still working on reproduction steps ### Expected behavior A more appropriate error should be thrown, and streaming should be cleanly terminated. ### Node.js version n/a ### @elastic/elasticsearch version 9 ### Operating system n/a ### Any other relevant environment information _No response_