[BUG] Upstream health check can get stuck after one async check failure

### Search before asking

- [x] I had searched in the [issues](https://github.com/apache/shenyu/issues) and found no similar issues.

### Apache ShenYu Component

shenyu-admin

### What happened

`UpstreamCheckService` keeps all submitted async checks in the shared `futures` list and clears it only after `CompletableFuture.allOf(...).join()` succeeds:

```java
private void waitFinish() {
CompletableFuture.allOf(futures.toArray(new CompletableFuture[0])).join();
futures.clear();
}
```

The scheduled task catches exceptions around `doCheck()`/`waitFinish()`:

```java
private void scheduled() {
try {
doCheck();
waitFinish();
} catch (Exception e) {
LOG.error("upstream scheduled check error", e);
}
}
```

However, `checkZombie(...)` submits `CompletableFuture.runAsync(() -> checkZombie0(...), invokeExecutor)` without an `exceptionally` handler, and the aggregation future inside `check(...)` can also fail if `updateHandler(...)` throws. For example, database writes, selector conversion, or event publishing inside `updateHandler(...)` can throw a runtime exception.

When any future completes exceptionally, `allOf(...).join()` throws `CompletionException`, so `futures.clear()` is skipped. The failed future remains in the shared list. Every later scheduled run includes the same already-failed future, so `join()` keeps throwing before the list can be cleared. This leaves the upstream health check loop stuck logging errors and retaining old futures instead of processing future health check results normally.

### Expected behavior

One failed async health-check/update operation should be logged and isolated to that cycle. The service should always clear completed futures for the cycle, for example by using `try/finally`, attaching exception handlers to every submitted future, or snapshotting and clearing the list before joining.

### How to reproduce

1. Enable HTTP register upstream health checking.
2. Have at least one upstream/zombie entry in `UPSTREAM_MAP` or `ZOMBIE_SET`.
3. Trigger a runtime exception from an async check path, for example make `updateSelectorHandler(...)` fail while processing a health status change.
4. `waitFinish()` throws from `CompletableFuture.allOf(...).join()` and skips `futures.clear()`.
5. Later scheduled runs keep joining the same failed future and cannot recover without restarting admin.

### Debug logs

_No response_

### Environment

Current `master` branch.

### Are you willing to submit a PR?

- [ ] Yes I am willing to submit a PR!

Start by locating UpstreamCheckService and reading scheduled(), waitFinish(), checkZombie(...), check(...), and updateHandler(...). Reproduce an async failure in the health-check path, then verify that the failure is logged and isolated to its cycle, completed futures are cleared, and later scheduled checks continue normally.

Written by the indexing model from the issue text.