v0.2.2

Added

• Sumstats loaders now accept genomatch-style summary-statistic headers in
  Python and MATLAB/Octave.
• MATLAB/Octave flat cache layouts now expose user-inspectable top-level
  variables for reference, sumstats, and annotation caches.
• Reference objects now cache deterministic a1_hash64 and a2_hash64 vectors
  for fast variant matching across Python and MATLAB/Octave.
• MATLAB/Octave can save thin reference caches with
  statgen.save_reference_cache(..., 'mode', 'thin'); thin caches store only
  metadata, base-pair positions, and allele hashes, improving
  load_reference_cache performance.

Fixed

• Improved MATLAB/Octave reference and sumstats loading performance.