First stable version

• Model expressions can be built for univariate, multi-samples
• NUTS and Metropolis-Hastings algorithm implemented
• NUTS optimized using STAN's logic
  • Adaptive up to diagonal precision matrix
  • Windowed approach as in STAN for warmup
  • Modified "state" variable (theta) to be momentum integration