There were some problems with the old str normalization. This fixes those
problems. Also, since it drills down on its own, it only needs to be
mapped over the blocks, not walked over the tree.
`normalize` from Text.Pandoc.Shared is more general. In tests, though,
it more than doubles the run time. `strNormalize` does less, but it does
what we need. This comment is added for future maintainability.
Use a function `stripSpaces`, instead of recursion. Makes it a bit
easier to read and mantain, and simplify normalizing DefinitionList,
which was left out the first time.
