Cleanup/getoncpu flush

This PR removes:

• RUNTIME_desc_getoncpu[_async]() functions
• MORSE_TASK_flush_[all|desc|data]( options, ... ) functions
• MORSE_Desc_Getoncpu()

And adds:

• RUNTIME_desc_flush( desc, sequence) that flushes and brings back to cpu asynchronously a full descriptor
• RUNTIME_data_flush( sequence, A, m, n) that flushes the piece of data A(m,n) and brings it back to the cpu if needed.
• RUNTIME_flush() that flushes all caches, but should be useless as we need to flush all descriptors used in the algorithms.
• MORSE_Desc_Flush( desc, sequence) to replace MORSE_Desc_Getoncpu( desc ) that was synchronous

All algorithms have been updated to use those new functions.

compute/pzbuild.c

@@ -30,7 +30,7 @@
 #include "control/common.h"
 
 #define A(m, n) A,  m,  n
-/***************************************************************************//**
+/*******************************************************************************
  *  Parallel tile matrix generation
  *
  *******************************************************************************
@@ -86,6 +86,5 @@ void morse_pzbuild( MORSE_enum uplo, MORSE_desc_t *A, void *user_data, void* use
     }
   }
 
-  MORSE_TASK_flush_desc( &options, uplo, A );
   RUNTIME_options_finalize( &options, morse);
 }