Use SSE3 in 32-bit x86 binaries

Pros:

• Seems to fix strange issue on AMD processors where the compiled binary has a chance to have 2x slower slope rendering in software mode
• Improves OpenGL performance on AMD processors
• Timedemo shows tiny performance increase on Intel Atom N270 on both renderers
• Timedemo shows tiny performance increase on Intel Core i7-4770HQ on software mode
  • Slopes get 20% faster on this CPU

Cons:

• CPUs older than 15 years will not run the executable.

The tests were done on the next branch.

CMake changes are untested, I don't have CMake.