restructure how selector types get inferred

[GGAlanSmithee]

We need to reconsider how to infer selector types, since there are a lot of possible combinations.

For example:

a#foo[bar="test"]

would have to be of type TypeId and AttributeExact, which won't work.

I'm thinking of doing some sort of bitflag system.

Will outline the idea after a bit of PoC:ing.

[GGAlanSmithee]

Also, there are another huge misstake, which is that the whole selector needs to be evaluated as one, which it doesn't right now. For example foo .bar should select all .bar elements that is a child of all foo elements, but now those gets split (I think) before even inferring their type.