Treat copied buffers as written only after completing the write in client

SYNC operation fsyncs only completed operations, so treating writes as "eligible
for fsync" before actually completing them is incorrect

It affected SCHEME=ec test_heal.sh (with immediate_commit=none) test - it was
flapping with lost writes - some non-fsynced writes were legitimately lost by
the OSD, but weren't repeated by the client

src/client/cluster_client.h

@@ -64,6 +64,7 @@ protected:
     unsigned bitmap_buf_size = 0;
     cluster_op_t *prev = NULL, *next = NULL;
     int prev_wait = 0;
+    uint64_t flush_id = 0;
     friend class cluster_client_t;
     friend class writeback_cache_t;
 };