Synchronize event in the CUDAProductBase destructor (#391)

Otherwise there are possibilities for weird races, e.g. combination of
non-ExternalWork producers, consumed-but-not-read CUDAProducts, CUDA
streams executing work later than expected (= on the next event).

CUDADataFormats/Vertex/BuildFile.xml

@@ -1,5 +1,6 @@
 <use name="cuda"/>
 <use name="rootcore"/>
+<use name="CUDADataFormats/Common"/>
 <use name="DataFormats/Common"/>
 <use name="HeterogeneousCore/CUDAUtilities"/>
 <use name="eigen"/>