Fallback Implementation is very slow #64
Description
The fallback implementation (especially for the Const versions which use a function pointer which is called per input byte, and so can't be optimized much by the compiler)
It appears possible for the compiler to optimize a struct with the bytes to search for equally well as a function, so it seems like the fallback could be implemented like that instead.
For backwards compat, we could keep the F generic parameter, or we could make a breaking change and remove it.
Goes best with #63, for the ability to build a const instance, and let the compiler see through it (which it can't do through a lazy_static/once_lock).