Check performance diff with manual hoisting of vectors

[CarlVerret]

Same here, hoist manually.

I have to admit that the code doesn't look as nice then, but from some previous work (in other parsers / validators) there was always a big enough boost to justify that change. But please measure it.

Basically it's similar to this example, compare the output from M1, and M2. You can see the load of the constant inside the loop in M1, but hoisted (manually) above the loop in M2.

Originally posted by @gfoidl in #91 (comment)