unicode.

Previously emoji like  (:man-man-boy:) were decoded as
:man::man::boy: () even though there is a specific emoji entry for
:man-man-boy:. The solution is to change how the regexp is constructed.
Instead of not setting a sort order for the emoji that go into the
regexp we use reverse alphabetical order to ensure that any emoji that
are inclusive of another (such as :man-man-boy: including :man) get
sorted BEFORE the emoji that they include.

We accomplish this by using a `SortedSet` and implementing `<=>` on `Emoji`

This allows a library user to easily customize the representation of an
emoji (mapping it to a custom image for example)

Without a pre-built lookup, we will look through all existing emojis (~1000) for non-existent emoji cheat codes. This is 30x faster than caching on demand in the presence of non-existent cheat codes, and roughly the same speed as caching on demand when there are no non-existent cheat codes.

This is a fix for #7.
In addition to the updated iterator, I've adjusted the numeric emoji to
use accurate unicode codepoints.

Fixes #1

A continuation of #3

In addition, this code rewrites encode to just defer to encode_io. This
is an attempt to fix #2, though I believe the lookahead code can and
should be refined.