In high dimension, gradient is almost our only hope. So methods like DRAM is useful probably only for some very specific model, and we did not see a big need of implementing it.
Of course contribution is always welcome 
It seems both DR and AM are possible to implement on top of the current Metropolis sampler (put the AM in the tuning stage, and DR in the sampler astep method). So I encourage you to start a PR/project, and we will be happy to help/review.