v2.4.0

• Add an explicit pinned column to chords.csv, keeping manual chord
  reservations independent from source frequency. Legacy blank-frequency
  pins remain compatible and migrate to explicit values on the next write.
• Recover orphaned alt forms when their base receives no chord; retain
  alt definitions across reruns and report only actual reachable coverage.
• Correct the standard keyboard's sixth column to the left hand so scissor
  scoring agrees with finger assignments and the displayed hand split.
• Measure learn timing from prompt activation, add isolated learn --recall
  practice with separate unassisted latency/speed histories, and migrate
  legacy progress without losing cards or daily quotas.
• Track learned mapping identity, preview generation with --dry-run, reserve
  learned mappings with --preserve-learned, and confirm intentional changes
  before retiring affected cards. Legacy cards bind once to current mappings.
• Weight chord assignments by uniquely served word families, summing each
  represented form's existing utility without changing source frequencies.
• Teach alt forms after their bases with independent FSRS cards and explicit
  modifier hints; drill/book now require each form's own demonstrated mastery.
• Add local, read-only analyze reports for personal-text coverage, family-aware
  learning recommendations, and observed recall versus an optional typing-speed
  estimate; document a separate manual mixed-typing passage comparison.
• Add read-only check diagnostics for dictionary/layout conflicts, configured
  export coverage and bindings, stale learning identities, and chord comfort.
• Add read-only difficult reports for lapses, longest unassisted recalls,
  and alt-slot evidence on verified current mappings. Diagnostic config
  validation no longer creates a missing dictionary directory.
• Removed dead find_all_combinations helper from utils.py and
  corrected the find_combinations docstring (results are ordered
  longest-first, not shortest-first).
• Cleaned up a variable-shadowing bug and an unused loop index in the
  Kanata output emitter.
• Added unit tests for find_combinations, validate_chords,
  TrainingOutput, the QMK/Kanata key-translation error paths, and the
  assigner's _split_into_tiers / _parse_freq helpers.
• Added gen.ignore_words config option. Words in this list stay in
  chords.csv but are skipped during chordgen gen scoring and chord
  assignment, so they won't consume chord keys or produce "unable to
  find options" warnings. Matching is case-insensitive.