Avoid a panic on task abort in Persist snapshotting

Cancellation can only happen when the Tokio runtime is being shut down, so it
shouldn't have meaningful production impact, but it does cause some noise
in testing. Use the prebuilt solution for it.

Author: bkirwi

src/persist-client/src/iter.rs

@@ -26,7 +26,7 @@ use futures_util::StreamExt;
 use futures_util::stream::FuturesUnordered;
 use itertools::Itertools;
 use mz_ore::soft_assert_eq_or_log;
-use mz_ore::task::JoinHandle;
+use mz_ore::task::{JoinHandle, JoinHandleExt};
 use mz_persist::indexed::encoding::BlobTraceUpdates;
 use mz_persist::location::Blob;
 use mz_persist::metrics::ColumnarMetrics;
@@ -597,9 +597,7 @@ where
 
                     let wrong_sort = data.run_meta.order != Some(RunOrder::Structured);
                     let fetch_result: anyhow::Result<FetchResult<T>> = match task.take() {
-                        Some(handle) => handle
-                            .await
-                            .unwrap_or_else(|join_err| Err(anyhow!(join_err))),
+                        Some(handle) => handle.wait_and_assert_finished().await,
                         None => {
                             data.clone()
                                 .fetch(