Which distinctions should be matched during normalization?

[xfq]

2.2.3 Unicode Normalization and character equivalence lists a lot of K decomposition results but does not specify which ones should match by default.

We should probably distinguish between canonical equivalence, compatibility equivalence, and the ignoring of diacritics, and to provide examples of both matching and non-matching cases.

In particular, NFKC should not be treated as a universal solution, as it may eliminate semantically significant distinctions.