This merges and cleans up some of the ugly copy/to from user code
which is required for the new fpr and vsx layout in the thread_struct.

Also fixes some hard coded buffer sizes and removes a redundant
fpr_flush_to_thread.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>