Add security considerations text.

[dlongley]

Addresses #70.

This PR removes the issue text in the Security Considerations section and adds text on dataset poisoning. The formal verification text is removed as a non-requirement for this section.

---

Preview | Diff

[iherman]

• I wonder whether it would be useful to suggest some values (based on current experiences) for the number of recursions and/or a reasonable timeout value for the suggested methods. Just to help implementers to start somewhere...
• (for a future editorial review) If dataset poisoning is the only security issue we refer to, then we may not want to have a separate subsection. It is all right to keep it as is for now, but just keep in mind when we publish the final version.

[yamdan]

Thank you @dlongley!

[dlongley]

@iherman,

I wonder whether it would be useful to suggest some values (based on current experiences) for the number of recursions and/or a reasonable timeout value for the suggested methods. Just to help implementers to start somewhere...

I thought about that -- but felt like it could vary depending on a number of factors to the point that saying something like the default timeout should be "on the order of a few milliseconds" or the max deep recursions should be "probably 1 for common use" felt wishy-washy.

I think what ecosystem a library was primarily written for (whether it's very general or more specific), whether it runs in a single thread or multiple threads (or has features to allow for other work scheduling), whether its target computing devices have X power or Y, and so on all impact the default values. The defaults are about what the predicted most common use cases will be. Ultimately, it's just about making the canonicalization algorithm run in some "reasonable amount of time" or some "reasonable about of compute" for their biggest user base -- which an implementer needs to figure out.

[philarcher]

Thanks for doing this Dave.

One thing: does canonize mean the same thing as canonicalize in US English? You've used 'canonize' a couple of times which, in my version of English, means the algorithm creates a Saint Dataset (i.e. to canonize is to make someone a saint). See https://www.merriam-webster.com/dictionary/canonize

[TallTed]

I believe we should use canonicalize in preference to canonize, throughout.

[dlongley]

Commenting to use suggestion feature.