Convert & to & as a Characters token

[willkg]

This fixes a problem in LinkifyFilter when using it with the Cleaner where
the Cleaner sets up the tokenizer to not consume entities. So character
entities end up in their own Entity tokens and Linkifyfilter can't match
links that cross token boundaries. If there's a &, then LinkifyFilter
won't match across that.

This fixes that by converting & to & in the sanitizer when it's pulling out
entities and putting them in separate Entity tokens. The & Characters tokens
will get merged by BleachSanitizerFilter.__iter__ and & will get converted
back to & in the serialier.

Fixes #422

[willkg]

These fiddly problems drive me crazy. This one took hours to work out. I think this is the best approach. I looked at a few others involving modifying the tokenizer, the parser, and wrapping LinkifyFilter in a merge_charaters kind of iterator. All those options had various problems from being hard to reason about or adding another performance hit or I couldn't make them work. Ugh.

One nice thing about this approach is that if we hit other character entities that need to be expanded, we can do that in this section and it'll "just work".

If you see anything better, I'm all ears.

[g-k]

r+ lgtm

[willkg]

Thank you!