Fix issues with x86 memcpy

Fix issues with x86 memcpy

The x86 memcpy() implementation did not mention its implicit output
registers ESI, EDI and ECX which might make this code miscompile when
the compiler uses the value of EDI for the return value *after* the 'rep
movsb' has completed. That would break the API of memcpy as this would
return 'dst+len' instead of 'dst'.

Fix this possible bug by removing the wrong comment and listing all
output registers as such (using dummy stack variables that get optimized
away).

Also the leading 'cld' is superflous as the ABI mandates the direction
flag to be cleared all the time when we're in C (see
<http://gcc.gnu.org/gcc-4.3/changes.html>) and we have no ASM call sites
that might require it to be cleared explicitly (SMM might come to mind,
but it clears the DF itself before passing control to the C part of the
SMI handler).

Last but not least fix the prototype to match the one from <string.h>.

Change-Id: I106422d41180c4ed876078cabb26b45e49f3fa93
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-on: http://review.coreboot.org/836
Tested-by: build bot (Jenkins)
Reviewed-by: Peter Stuge <peter@stuge.se>