Integrate dataflush in QR/LQ algorithms

Add RUNTIME_data_flush calls in all QR/LQ algorithms to start working on the distributed implementation of those algorithms on top of StarPU.

Replace !61 (closed) that was coming from rboucher/chameleon and on which we don't have the rights to enable the runners.

compute/pzunglq.c

@@ -116,6 +116,8 @@ void morse_pzunglq(MORSE_desc_t *A, MORSE_desc_t *Q, MORSE_desc_t *T, MORSE_desc
                     Q(m, k), ldqm,
                     Q(m, n), ldqm);
             }
+            MORSE_TASK_dataflush( &options, A(k, n) );
+            MORSE_TASK_dataflush( &options, T(k, n) );
         }
 #if defined(CHAMELEON_COPY_DIAG)
         MORSE_TASK_zlacpy(
@@ -147,6 +149,9 @@ void morse_pzunglq(MORSE_desc_t *A, MORSE_desc_t *Q, MORSE_desc_t *T, MORSE_desc
                 Q(m, k), ldqm);
         }
 
+        MORSE_TASK_dataflush( &options, D(k)    );
+        MORSE_TASK_dataflush( &options, T(k, k) );
+
         RUNTIME_iteration_pop(morse);
     }