prize18-6  Exercise 2: Composition Does Not Change Clusters
===========================================================

Predictions: three Hangul syllables are 3 code points in NFC and 8 in NFD (the three syllables split into 3 + 3 + 2 jamo), but 3 clusters either way.
"a", CR LF, "b" is 4 code points and 3 clusters. A letter followed by two combining marks is one cluster; a combining mark before any letter has nothing
to attach to, so it is a cluster of its own, and the letter after it another: 2 clusters.

ex2.py (needs the third-party regex module):

  import regex
  import unicodedata as ud

  hangul = "\ud55c\uad6d\uc5b4"        # three syllables
  for form in ["NFC", "NFD"]:
      s = ud.normalize(form, hangul)
      print(form, len(s), len(regex.findall(r"\X", s)))

  crlf = "a\r\nb"
  print(len(crlf), len(regex.findall(r"\X", crlf)))
  print(len(regex.findall(r"\X", "e\u0301\u0323")), len(regex.findall(r"\X", "\u0301e")))

Real output (Python 3.14):

  NFC 3 3
  NFD 8 3
  4 3
  1 2

All as predicted. The key point is that the cluster rules are written in terms of properties of the code points, so a precomposed syllable and its
decomposed spelling land in the same place; normalization changes code point counts but not what the reader sees.