Document expected performance

[xaviershay]

Forking from #134.

Given we are making performance claims ("At the same time, they remain CPU and memory-efficient by sharing between copies.") we should include expected operation cost in the YARD docs (O(1), O(n), etc).

At the same time, proving an "implementation details" section for each collection type would be useful to future committers. Something like "Hash is implemented using a persistent Trie, as outlined in . It provides the following performance guarantees: ..."

If anyone was able to brain dump details into this issue, I'm happy to clean them up and amend docs.