Add normalization of combining dakuten and handakuten to ja preprocessors

[Kuuuube]

Fixes #1116
Possibly optimization overkill. The way I wrote it is 6x faster than the dead simple way of regex + normalize:

text.replace(/.(\u3099|\u309A)/g, (match) => {
    return match.normalize();
});

But idk if it's worth the code (probably) looking like witchcraft to someone else that sees it.

Also probably needs some cases added to tests.

[Casheeew]

The code looks pretty clear to me, and considering dakutens appear quite often faster probably brings no harm

[Kuuuube]

Added tests. This should be good to go.

Resolved