shrinking
the size of sequences in a preference-order preserving fashion.

Finally, note that it isn't strictly necessary to use an [`Extractor`]. Namely,
an `Extractor` only uses public APIs of the [`Seq`] and [`Literal`] types,
so it is possible to implement your own extractor. For example, for n-grams
or "inner" literals (i.e., not prefix or suffix literals). The `Extractor`
is mostly responsible for the case analysis over `Hir` expressions. Much of
the "trickier" parts are how to combine literal sequences, and that is all
implemented on [`Seq`].