[WIP] Compress to a single-pass

[botandrose]

This PR attempts to improve the algorithm further by removing the second pass. Rather than moving nodes to the pantry, and then back in later, it moves them to the correct place all within the first (and only) pass.

Note that this is an evolution of the v0.4.0 two-pass algorithm, and is behind the same option, so default mode still behaves like v0.3.0, and one must still enable the new algorithm via the twoPass option (which has become a bit of a misnomer).

Benefits:

1. No more need for the abstraction-leaking beforeNodePantried callback. If you are doing interesting things with any of the other hooks, e.g. data-turbo-permanent, you'd often need to prevent the pantrying process from flattening the node trees that entered it, and thus you'd need to know how the internals of the pantrying process worked, and that there was a pantrying process at all, for that matter. This PR removes the hook, and the need for it.
2. Hooks are called in the order one would expect, and with the arguments one would expect. Two-pass mode resulted in minor changes to both in v0.4.0. This PR results in hook behavior that is more correct than both v0.3.0 and v0.4.0.
3. Many minor suboptimal situations have become optimal. For more details, see the test changes in the diff.
4. Less DOM operations than v0.4.0, so maybe more performant?

Finally, as far as I can tell, there are no downsides to this new approach, but I welcome any skeptical scrutiny to make this as good as it can be! Can we find any regressions from v0.3.0 or v0.4.0?

P.S. First two commits are from #83 , which is a bugfix and should be evaluated separately from the outcome of this. I'll rebase once that gets merged.

[botandrose]

Closing in favor of #93