Wrong decompilation of "rep stos"

Subject: https://github.com/rfalke/decompiler-subjects/blob/master/from_holdec/i386_memset_rep_stos/ia32_elf/subject.exe

Issues:
* `rep stosw` generates the same decompiler output as `rep stowd`. See functions `i386_memset_forward_w()` and `i386_memset_forward_l()`. Maybe one of the `__asm_rep_stosd_memset()` should be `__asm_rep_stosW_memset()`?
* retdec doesn't take into account the value of the direction flag. While it is a convention that it is cleared it doesn't have to be. The functions with `backward` or `both` in name show this.

Start with the referenced subject.exe and compare the decompiler output for rep stosw and rep stosd in i386_memset_forward_w() and i386_memset_forward_l(). Then inspect the backward and both functions for direction-flag handling. Done means word and double-word stores decompile distinctly and the direction flag is accounted for rather than assumed clear.

Written by the indexing model from the issue text.