Adding FMA and AVX2 config switches
Most modern x86 processors have FMA and AVX2, which OPUS could leverage for better performance.
This commit replaces existing AVX flag to a new flag guarding all AVX, FMA, and AVX2; mainly because:
- AVX-optimized FP code usually comes with opportunities for FMA;
- CPUs support FMA but not AVX2 are rare (AMD Bulldozer Gen 2 and 3);
- No CPU supports AVX2 but not FMA;
- MSVC's /arch:AVX2 controls both FMA and AVX2.