Protect runtime_data_flush to be able to easily disable it

RUNTIME_data_flush is now hidden at the compute level behind chameleon_data_flush.

This is done for two reasons:

• You don't want to flush piece of data with recursive tasks, so it will be simpler to disable it in the request associated to the recursive task (@tmorin)
• when submitting multiple async algorithm, you may not want to flush data between two algorithms. So you enable the flush only in the last request of the sequence.