Add faster and more accurate WindowedSinc interpolators. (#375)

* Try speeding up resamplers with SIMD.

* Speed up all resamplers by ~25%.

* temp, kinda works

* Reworking logic in steps

* temp

* It works for everything but WindowedSinc

* Working and fast for Linear and 0OH

* working

* god I hate this, givin gup

* Implement faster WindowedSinc interpolators.

* Disable mostly-fast-math.

* Update default resampler quality to WindowedSinc32.

* Really, clang-format?

* Remove extra speed tests.

* More cleanup.

* Update version.