Broken Hamiltonian MCMC with 2D Gaussian distribution

[rccreswell]

Based on discussion in #3 , try breaking the Hamiltonian sampler.

• Break the acceptance step with some probability
• Introduce an error into the calculation of the gradient