Refactor offset-allocator to improve readability

[patowen]

A few months ago, as part of my work on Hypermine, I was looking into this crate as something to potentially use to manage allocations in Vulkan. While evaluating this crate, I studied the code carefully to understand how it worked, and while doing so, I made some rather extensive refactors to make it easier for me to understand this code. Eventually, I was happy enough with the refactors that I was in favor of vendoring this dependency in Hypermine so that I could take advantage of these refactors.

However, it was suggested that I look into trying to contribute my refactor upstream. I was initially hesitant to do so because this crate was meant to be a port of a C++ package, which itself is considered an "early one weekend prototype", and it seemed like this crate is meant to be as faithful to that C++ package as well.

Regardless, I'm submitting this draft PR in case you are interested in seeing what I have, in case such a refactor is worthwhile to have upstream. I've rebased the commit history to hopefully ease reivew.

This PR makes the following changes (which have been separated out by commit):

• Add and update documentation to better explain how the allocator works
• Create various abstractions, including SmallFloat, BinsMap, and NodeSlab
• Move some code into separate modules, including NodeIndex, BinsMap, and NodeSlab
• Add various bugfixes and tweaks

Please let me know if you have any further questions about this PR. I promise that I did not use an LLM to write any of the code in this PR.

[patowen]

I've updated this PR so that it's actually readable, as well as removing some of the more controversial changes (such as adding the slab dependency and removing the nonmax dependency).