mutation: 72 stale entries re-pointed, 15 that stopped compiling fixed

The first full run on CI named every entry that could not prove anything. Two
kinds are fixed here, and each was run again locally afterwards.

- 72 entries whose pattern was gone. Most of the code they break was rewritten
  in place - messages moved from fmt.Sprintf to core.Says, palette values were
  recomputed - and three moved to another file. Each now breaks the same
  behaviour where it lives today. Where a pattern carried a number that changes
  with the code, the entry takes a shape that does not: the README count gains a
  leading digit, the file ceiling gains "1 +", the Polish catalogue is addressed
  by key rather than by hash. 65 caught, 6 not caught (left for the next pull
  request, with what was found), 1 unjudged here.
- 15 entries whose mutant no longer compiled - a variable the mutant orphaned, a
  string where a core.Said is now asked for, a method that is gone. All 15
  caught. One needed a second aim: the guard looks for a title carrying the
  label, so the break now gives a fragment that title.

A BROKEN verdict now carries the compiler's first complaint. On CI the verdict
was all there was, and these fifteen had to be built again by hand to learn why.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Author: donislawdev