Avoid calling OrderedDict unnecessarily

[gsnedders]

Making normalizeToken only call OrderedDict when we have more than one attribute leads to a big jump in parsing speed when parsing the HTML spec in some quick testing. Need to make sure we never append to that for it to be a safe optimisation, though.

[gsnedders]

Need to make sure we never append to that for it to be a safe optimisation, though.

We do, and especially with dict by ordered from Python 3.7 (and #493 makes us use it then), the cost here is small. I wonder if special-casing only the zero-attribute case is worth the pain it would cause? And hey, we have benchmarks now.