0.4.1

Fix a packaging error on AVX builds caused by an old version of stdsimd

0.4.0

- 4x faster vectorized encoding
- 2x faster vectorized decoding
- Now 2-4x as fast as base64

We've done all we can do on one core - thankfully, this is 2017 and processors
have tens of cores. Next up: fearless concurrency!

0.3.0

This release includes

- 4x faster scalar decode
- 2x faster scalar encode
- SIMD acceleration on nightly {feature = "simd"}
- Removal of fast mode
- Fix egregious decoding bugs
- Now /actually/ faster than base64

This release isn't compatible with 0.2.0 because of bugs in 0.2.0 and the
removal of fast mode, but the encoding algorithm did not change.

I have around 100 lines of explicit AVX2 code laying around for 0.4.0 - Let's
make this thing not just faster than base64, but 10x faster than base64.

0.2.0

This release contains:
- Order-of-magnitude performance improvements
- Total incompatibility with 0.1.0
- Stable compiler support
- Graceful error handling