Move simd under a feature flag, use fast CharLookup if not enabled

[Wonshtrum]

SIMD makes parsers 2 times faster in "pre-body phase". But it uses x86-specific instructions, making it not compilable on other architectures (see #4). This commit:

• puts all x86 instructions in "_simd" suffixed functions
• declares non-suffixed functions that use "simple" (but fast) char lookups
• declares "_fast" suffixed functions that use the fastest available functions (simd if enabled, char lookup as a fallback)

This also allows for other architecture-specific optimizations by declaring other suffixed functions and adding them to the "_fast" suffixed ones with a feature flag.