This PR changes how we calculate the trait/implements hierarchy.

Currently an impl Trait for T block makes T directly implement both Trait and any supertraits, SuperTrait, of Trait. However, in order for this impl block to be valid, it must already be the case that T implements SuperTrait by some other impl block. Hence, the only new information from an impl block is the implementation of the specific target trait.

This PR changes how we handle impl blocks when calculating the trait/implements hierarchy, s.t. impl blocks only makes a type implement the specific trait in the impl block.

This restriction is also used in AssocFunctionType where the restriction removes some spurious call targets.

The DCA report shows a small speedup and decent reductions in "Path resolution inconsistencies" and "Nodes With Type At Length Limit". There is a small increase in "Unknown expression types". I think this increase looks reasonable when compared to the decrease in "Path resolution inconsistencies". I also did a quick and small spot check on neon, and the lost types all looked like spurious types.