fix: add ScopeGuard to wait async tasks before early return in Clean()

When GetUsedFiles() returns an error (e.g. index manifest not supported),
Clean() would return immediately while thread pool tasks submitted via
Via(executor_.get(), ...) were still running. These tasks capture 'this'
and access members like fs_, causing use-after-free when the
OrphanFilesCleanerImpl is destroyed shortly after.

Add a ScopeGuard that calls CollectAll(file_statuses_futures) to ensure
all submitted async tasks complete before the function returns, preventing
the intermittent segmentation fault.

Author: zjw1111

src/paimon/common/executor/future.h

@@ -96,7 +96,9 @@ std::vector<T> CollectAll(std::vector<std::future<T>>& futures) {
     std::vector<T> results;
     results.reserve(futures.size());  // Reserve space to avoid reallocation.
     for (auto& future : futures) {
-        results.push_back(future.get());  // Wait for each future and collect the result.
+        if (future.valid()) {
+            results.push_back(future.get());
+        }
     }
 
     return results;

src/paimon/core/operation/orphan_files_cleaner_impl.cpp

@@ -97,6 +97,8 @@ Result<std::set<std::string>> OrphanFilesCleanerImpl::Clean() {
     }
     PAIMON_ASSIGN_OR_RAISE(std::set<std::string> all_dirs, ListPaimonFileDirs());
     std::vector<std::future<std::vector<std::unique_ptr<FileStatus>>>> file_statuses_futures;
+    ScopeGuard file_statuses_guard(
+        [&file_statuses_futures]() { CollectAll(file_statuses_futures); });
     for (const auto& dir : all_dirs) {
         file_statuses_futures.push_back(
             Via(executor_.get(), [this, dir] { return TryBestListingDirs(dir); }));