@iarna/toml

Another 18% speed boost on our overall benchmarks! This time it came from
switching from string comparisons to integer by converting each character to
its respective code point. This also necessitated rewriting the boolean
parser to actually parse character-by-character as it should. End-of-stream
is now marked with a numeric value outside of the Unicode range, rather than
a Symbol, meaning that the parser's char property is now monomorphic.

Bug fix, previously, 'abc''def''' was accepted (as the value: abcdef).
Now it will correctly raise an error.

Spec tests now run against bombadil as well (it fails some, which is unsurprising
given its incomplete state).