The tests check for consistency between Ord, Eq and Hash

Hashing does not have to use the whole Components parsing machinery because we only need it to match the
normalizations that Components does.

* stripping redundant separators -> skipping separators
* stripping redundant '.' directories -> skipping '.' following after a separator

That's all it takes.

And instead of hashing individual slices for each component we feed the bytes directly into the hasher which avoids
hashing the length of each component in addition to its contents.

…f trying to hash the raw bytes

This should have 0 performance overhead on unix since Prefix is always None.