Fix: undo commited local CUDA compute capability choice

Update README.md

Tweaks

First working SIMD version with basic AVX2 support (runtime use is de…

…cided at compile time)

Fixes issues in md5birthdaysearch:

- fixed datarace in birthday collision verification
- replaced hardcoded fixed CUDA buffer size with formula
- use 2x more CUDA blocks by default for better efficiency

create_release.sh: fixed issue with recursive make calls and specifyi…

…ng Makefile.tmp

Added fastcoll 1.0.0.5-1 to hashclash.