- 10/26: version 0.1.6
  * build-time trimming of riskiness, sharp-YUV, search, adaptive quantization,
    metadata and C fallbacks: library down to 37 KB (everything off, -Oz)
- 10/26: version 0.1.5
  * fast mode, -fast (EncoderParam::fast_mode): 15-30% faster
- 10/26: version 0.1.4
  * progressive JPEG encoding (partial)
  * restart markers (DRI/RST0-7) and multi-threaded encoding using them:
    parallel scans, adaptive-quantization histograms, multi-pass search and
    sharp-YUV conversion
  * new quantization options: fast RDO (-rdo), adaptive bias
  * progressive-aware SjpegRiskiness() with early exit (~2.6x faster)
  * native RGBA/BGRA input paths (zero-copy)
  * AVX2 functions: fdct, quantization, histogram, color conversion, sharp-YUV, riskiness
  * faster SSE2/SSSE3/NEON: planar conversion, quantization (branch-free
    run-length scan), histogram, trellis, fdct, JPEG scoring
  * MSVC x86/ARM64 SIMD support, thread-safe lazy init, overflow fixes
  * pkg-config file, unit-test binary, 'check' CMake target, CI
- 01/20: add YUV400 (grayscale) support
- 06/23: fix search-to-target-size mode
- 03/18: PSNR/size search hooks, SetMinQuantization(), NEON QuantizeError
- 11/17: initial release version 0.1
