Fixed the swallowed IOStatus in Compaction Job introduced in PR 7718 (#7838)

Summary:
The IOStatus of TableBuilder is returned by copy the io status from builder->io_status(). pr https://github.com/facebook/rocksdb/issues/7718 swallowed the io status and it will cause the write IO error become non-retryable and no auto resume logic will handle it. Roll back to previous implementation.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/7838

Test Plan: make check

Reviewed By: ajkr

Differential Revision: D25795387

Pulled By: zhichao-cao

fbshipit-source-id: bc35e69e0b71aa4148a6ed76f073357041b8e372

db/compaction/compaction_job.cc

@@ -1449,8 +1449,9 @@ Status CompactionJob::FinishCompactionOutputFile(
   } else {
     sub_compact->builder->Abandon();
   }
+  IOStatus io_s = sub_compact->builder->io_status();
   if (s.ok()) {
-    s = sub_compact->builder->io_status();
+    s = io_s;
   }
   const uint64_t current_bytes = sub_compact->builder->FileSize();
   if (s.ok()) {
@@ -1461,7 +1462,6 @@ Status CompactionJob::FinishCompactionOutputFile(
   sub_compact->total_bytes += current_bytes;
 
   // Finish and check for file errors
-  IOStatus io_s;
   if (s.ok()) {
     StopWatch sw(env_, stats_, COMPACTION_OUTFILE_SYNC_MICROS);
     io_s = sub_compact->outfile->Sync(db_options_.use_fsync);